Collaborative Mission Planning, Autonomy and Control ... · Collaborative Mission Planning, Autonomy and Control Technology (CoMPACT) for Unmanned Surface Vehicles ∗ Joshua Redding†,

Collaborative Mission Planning, Autonomy and Control Technology

(CoMPACT) for Unmanned Surface Vehicles ∗

Joshua Redding†, Jayesh Amin‡, Jovan D. Boskovic§, and Joseph Jackson¶

Scientific Systems Company, Inc.

In this paper, a Collaborative Mission Planning, Autonomy and Control Technology (CoMPACT) is pre-

sented for enhancing Unmanned Surface Vehicle (USV) capabilities in a maritime setting. CoMPACT uses

a Finite State Machine (FSM) structure, can entail an array of communication architectures, and is demon-

strated here for USVs under anti-submarine warfare (ASW), mine countermeasures (MCM), and other search

and inspection tasks. FSM decision rules enable event-based or information-based transitions to govern the

behavior of the autonomous agents. As part of this application and testing of CoMPACT, a study of innovative

multi-agent cooperative search patterns conducive to reconfiguration (e.g. addition or deletion of agents from

the search team due to reassignment or failure) is included with the simulation results. CoMPACT features

several levels of autonomy, from completely tele-operated to fully autonomous mission completion, designed to

reduce operator fatigue when several USVs are issued routine tasks.

I. Introduction

A growing percentage of future conflicts are expected to take place in the littoral regions around the globe against

adversaries who possess increasingly effective weapon systems. Unmanned air, ground, underwater, and surface sys-tems present an effective paradigm for warfighting in these theaters of operation during missions too hazardous for

manned vehicles. Hence unmanned systems are critical components of the future naval forces. While significant re-

search and development has been performed on multiple unmanned underwater vehicles (UUVs)1 and unmanned aerialvehicles (UAVs),2–4 relatively little has examined coordination and cooperative control of unmanned surface vehicles

(USVs).

A USV offers many benefits to the Fleet. The first, and most obvious, is ensuring the sailors’ safety. A USV

can be deployed in waters where it is unacceptable to send a manned vessel, including high threat environments or

areas contaminated by nuclear, biological, or chemical agents. A USV could also remain on station for extremely longperiods of time (up to several weeks) without resupply or human intervention. Such a capability could allow for long-

term anti-submarine warfare (ASW), mine countermeasures (MCM), and Sea Base protection operations in areas of the

world where future conflicts are likely. Additionally, a USV squadron could be deployed in advance of a carrier battlegroup or amphibious ready group to sanitize the area of potential threats and assure access for troops. Furthermore,

USVs have large payload capacity, allowing for a multi-function mission package to be deployed with each USV. A

single USV could simultaneously conduct many operations.

Operating on the water’s surface, a USV can use conventional power sources, such as diesel or gas turbines rather

than relying on more limited power supplies, such as batteries or fuel cells. USVs are also unique among unmannedsystems since they may communicate in all three mediums of interest—undersea, air, and space—relaying information

from submerged assets (submarines, UUVs) to and from any combination of surface vessels, aircraft, or satellites.

A. Motivation for Collaborative Autonomy for USVs

The current concept of operation that requires one or more operators per USV must be extended to broaden thecapabilities of USVs through operational concepts which take advantage of concurrent multiple vehicle deployment.

∗This research was supported by Office of Naval Research contract No. N00014-07-M-0392 to Scientific Systems Company.†PhD Candidate at Massachusetts Institute of Technology, currently on leave from SSCI, AIAA Member, [email protected]‡Lead Research Engineer, 500 W. Cummings Park, Suite 3000, Woburn, MA, [email protected]§Principal Research Engineer and Group Leader, 500 W. Cummings Park, Ste 3000, Woburn, MA, AIAA Senior Member, [email protected]¶Research Engineer, 500 W. Cummings Park, Suite 3000, Woburn, MA, AIAA Member, [email protected]

1 of 23

American Institute of Aeronautics and Astronautics

AIAA Guidance, Navigation, and Control Conference10 - 13 August 2009, Chicago, Illinois

AIAA 2009-5774

Copyright © 2009 by Scientific Systems Company, Inc. Published by the American Institute of Aeronautics and Astronautics, Inc., with permission.

Proceeding with multiple USV operations under the current vehicle command and control approach would require

unacceptable numbers of operators and potentially result in diminishing returns if coordination techniques and tools

are not developed.

When USVs operate cooperatively, these operations inherently become more complex than the one vehicle, single

sensor, single search area operations. As multiple mission functionality is implemented, the system tools to optimizemulti-vehicle/multi-task performance are necessary. Thus cooperative behavior amongst USVs, supporting USVs as a

“Force Multiplier”, will focus on: (i) Reducing individual operator workload; (ii) Reducing the number of operators

required; (iii) Improving the responsiveness (and decreasing reaction times) of deployed USVs; and (iv) Ensuringmaintenance of overall USV search coverage, while optimizing interrogation/inspection capability. Since USV utility

is directly related to number of operators required, the force multiplier potential of USVs can be attained only if the

collaborative autonomy design achieves multiple vehicles per operator rather than multiple operators per vessel.

B. Current results in the area of cooperating unmanned vehicles

Many studies of cooperative control of unmanned vehicles have been published. While most of the results apply

to unmanned aerial vehicles (UAV) and, to a lesser extent, to unmanned underwater vehicles (UUV), many proposedtechniques are sufficiently general and hence applicable to surface vehicles as well.

A key problem in the area of coordinated control of multi-agent systems is the assignment problem that represents

a well-known optimization problem. In its simplest form, it is to assign items (jobs, missiles/strike vehicles) to otheritems (machines, targets).7 The solution to the multi-agent assignment problem has been attempted using numerous

techniques including task-decomposition8 and auction-based tasking.9 Although algorithms for solving the assignment

problem subject to a performance criterion exist, the problem is NP-hard (i.e. it cannot be solved in polynomialtime); thus, heuristic techniques are often used.10 In the context of coordinated control of multi-agent systems, the

assignment problem is commonly coupled with that of trajectory optimization, making the overall problem highly

complex. Numerous approaches have been proposed for UAV coordination, from those based on rigidly defined desiredrelative position vectors and trajectory optimization techniques,2, 4 to approaches based on loosely defined inter-vehicle

interactions such as through artificial potential field methods.5

The coordinated control problem can be loosely defined as that where multiple agents interact and coordinate

their dynamic behaviors in order to accomplish a set of tasks that are necessary to achieve mission goals. Several

issues arise in the context of multi-UAV applications such as aerial surveillance and tracking, collision and obstacleavoidance, formation reconfiguration, high-level control, and hardware and communications.3 Most of these problems

also arise in the context of USVs, where a typical mission consists of several highly coupled complex tasks such as

area search, vehicle-to-target assignment, removal/addition of vehicles from/to the platoon, platoon reconfiguration,and target pursuit and inspection. All these tasks need to be accomplished under spatial, timing and vehicle constraints,

often in a highly dynamic setting. Hence typical behavior-based approaches5 may not be best suited for coordination of

the above tasks under dynamic constraints. On the other hand, optimization-based techniques,4 based onMixed-IntegerLinear Programming (MILP) may result in unbounded computing times and may not be implementable in real time.

Heuristic approaches may alleviate this issue, but their robustness to uncertainty needs to be demonstrated.

II. Autonomy Approach and Design Specifications

Autonomy is defined by the Merriam-Webster dictionary as “the quality or state of being self-governing” or alter-

natively as “the state of being free from the control or power of another.” In the context of military scenarios involving

multiple unmanned vehicles, elimination of any and all human control and supervision is not yet feasible or evendesirable at this stage. However, in accordance with the guidelines of “crawl-walk-run” as laid out in the USV and

UUVMaster Plans,11, 12 designing systems that implement increasing levels of autonomy provides a transitional frame-

work system that can take us from the existing multiple-operators-per-vehicle setup to multiple-vehicles-per-operator.Such improvements provide great benefits as a force multiplier while also reducing risks to the sailors’ lives amid

dirty/dull/dangerous missions in a complex littoral environment. In order to achieve this vision, we have developed

a Collaborative Mission Planning, Autonomy and Control Technology (CoMPACT) system for multiple USVs. TheCoMPACT system is shown schematically in Figure 1.

Objectives of CoMPACT are to 1) provide a framework for autonomous collaboration, coordination & control, con-flict resolution, re-planning, and assignment of heterogeneous USVs; 2) decrease operator workload using adjustable

degrees of autonomy; 3) be modular, adaptable and capable of intelligent operation in a multitude of scenarios; 4)

extend to a variety of different communication schemes and agent-level collaboration definitions.

2 of 23


Autonomous ControlSystem

Vehicle Sensors

Autonomous USV

CoMPACT


Vehicle Sensors

Autonomous USV

Communication Network


Vehicle Sensors

Autonomous USV

Awareness SensorsGlobal Situational

OperatorCommand Center

Mission Specifications

Figure 1. Structure of the Collaborative Mission Planning, Autonomy and Control Technology (CoMPACT) system for multiple USVs

The implementation of CoMPACT presented here utilizes event-based task transitions, efficient obstacle repre-

sentation, and fast dynamic path (re)planning. Ensuing sections address the autonomy architecture design, CommonOperating Pictures (COPs), and how CoMPACT accommodates various levels of autonomy as required by the operator

or mission scenario.

Mission specifications are generated by the command center that take into account vehicle availability, suitability

for the mission or task, and dynamic constraints (maximum velocity, sensor range and field of view, maneuverability,

etc.) The CoMPACT system receives high-level commands (“search the area”, “inspect a suspicious vessel”), anddecomposes them into a set of coordinated tasks. For instance, for the area search, it assigns each vehicle to a part of

the search area, and monitors vehicle assignment for target pursuit and inspection. To avoid a single point of failure,

the assignment algorithm is decentralized so that the vehicles calculate their own corresponding performance index

and communicate it to the neighboring vehicles. The vehicle with the best performance index is chosen for the task.

The proposed system is sufficiently general and covers the spectrum of control architectures, such as: (i) Fully

centralized, where the Coordinated Control System (CCS) performs all the task assignments, and coordinates theirexecution; (ii) An architecture with a Local Leader, where the CCS assigns higher-level tasks to smaller platoons that

each have their local leader responsible for task execution; and (iii) Highly decentralized, where USVs perform taskassignment on their own by calculating the performance indices and communicating them to the neighboring vehicles;

For instance, a USV is assigned a loosely-defined task (e.g. “go and inspect the vessel at coordinates (x, y)”, “when

done return to the search area”), carries out pursuit and inspection, returns to the search area, carries out re-tasking forsearch while avoiding pop-up threats, such as mines, and collisions with friendly vessels.

The CoMPACT system is designed to efficiently handle the following diverse missions and tasks:

• Area search with a platoon of USVs;

• Vehicle-to-target assignment, and removal of vehicles from the platoon to pursue, inspect and neutralize mines

and/or suspicious vessels;

• Trajectory generation with dynamic obstacle avoidance for intercept, inspection and data collection;

• Autonomous reconfiguration of remaining vessels to continue the search;

• Return of the pursuit & inspection vehicles to the platoon;

• Platoon re-tasking to continue the search with additional vehicles; and

• Replanning under failures and faults.

Specifically, this paper presents results of the implementation of the CoMPACT system under three scenarios.

Documented here are missions such as reconfigurable coordinated search, where multiple agents are tasked to surveya region in a cooperative manner subject to configuration changes (Figure 2), and MCM and ASW missions, where

teams are comprised of members with specific payloads and sequential roles in the mission (Figure 3).

3 of 23


Figure 2. Reconfigurable Search Scenario

Figure 3. MCM and ASW Scenario

4 of 23


III. Communication Architectures

An important driving factor for autonomy design are the communication constraints imposed by the mission sce-nario. A domestic urban unmanned surveillance operation would have available almost unlimited reliable communi-

cations along with sufficient human supervisors. On the other hand, a blockade mission in a remote hostile littoral

environment would have extremely unreliable and constrained communications as well as requirements to minimizethe risk to human lives. This range of communication constraints and available topologies drives the choice between

centralized or decentralized autonomy designs. Mission scenarios we have addressed in this project mainly require

a decentralized approach. However, it is also important to have a design that can easily be adapted to a centralizedapproach when necessary or appropriate.

CoMPACT implements an Intelligent Agent (IA) structure, as illustrated in Figure 1, that does not hinge on anyone collaboration structure to operate. The autonomy can be adapted to a number of architecture designs based on

the mission. Constraints that contribute to architecture selection include communications bandwidth, communications

range, on-board processing limitations, robustness requirements, and typical mission profiles. The following sectiondescribes the various communication architectures and topologies for a collaborative distributed system.

A. Centralized or Global Leader Architecture

A common method for control of multi-agent systems is the centralized approach that may be implemented either

through a central command center (LCS or remote ground station), or through a single USV designated as the globalleader. Technically, these are similar in the sense that all the information is gathered at a single point for decision mak-

ing and task allocation for the individual USVs by the decision maker. In the Global Leader collaboration architecture,

Figure 4, the vehicles rely upon one USV to gather all of the information regarding USV states and capabilities, thenmake assignment decisions based upon the respondants status.

Figure 4. The Global Leader architecture requires all assignments be issued from a central agent.

Considerations for the centralized approach include requirements for the Leader to have enough bandwidth to

receive updates from all USVs in the platoon, have sufficient broadcast range to communicate with the farthest USV,and have sufficient processing power to arrive at the best assignment configurations quickly. It is also important to note

that, as the number of USVs in the operations increases, the communication and supervision requirements grow very

rapidly making it hard to scale. However, having a Global Leader make decisions removes uncertainty about whichagents are best suited to perform certain tasks, and likely can provide global optimal allocations compared to potential

local optimal allocations in the decentralized implementations. Likewise, a Global Leader structure will not result in

conflicting actions among the agents. The biggest disadvantage of a centralized approach is its vulnerability to a singlepoint of failure. Any loss of the leader or its communication capabilities may render all the other assets unusable.

B. Local Leader Architecture

The next level of decentralization is achieved through clustering of vehicles. The idea of a Local Leader, Figure 5(a),

is that decisions can be more efficient and robust when no one agent is responsible for issuing all the assignments. Forexample, if the platoons are organized by geographical location and a need arises for a USV to inspect an unknown

5 of 23


target at a certain location, the Local Leader of the platoon closest to the unknown threat may be assigned the task. The

Local Leader may then request status from each USV within its platoon, and the agents will be given assignments after

the responses are gathered.

This decision-making structure, when scaled properly, can be faster at arriving at efficient assignments, and tends

to be more robust to single-point failures, since several agents are cabable of distributing the decision-making process.Furthermore, the local decision making can be handed-off to another USV in the platoon if the Local Leader drops out

of the communications link. Local Leaders enable a longer-range communications link since not all agents need to be

within the communications range of the LCS.

C. Decentralized Architecture

A completely decentralized architecture moves the decision making ability to the individual vehicles (Figure 5(b)).

However, to be able to make collaborative decision, the vehicles now need to have information about other vehicles

and assets that are available. This is discussed in detail below.

1. Common Operating Picture, Common Relevant Operating Picture

Truly decentralized architectures necessitate the concept of a Common Operating Picture (COP). COP is a database

of the locations, capabilities and status of all the other assets in the field. COP also includes other world informationlike missions parameters, target data, terrain or bathymetry data, and so on, which are necessary for performing the

mission. When individual vehicles need to make a decision as to who needs to respond to an event, they all use the data

from their own copy of the COP. If all the vehicles use the same algorithm, then barring inconsistencies in the COPdata, they should all arrive at the same decision without the need for negotiations or centralized supervision. If COP

is to be synchronized by sharing data every time a decision is required, most advantages of the decentralized approach

are lost through these increased communications and delays.

One approach is to use the low activity periods in the field, when no significant events are occurring, to periodically

share information between the vehicles. Additionally, not every vehicle needs to store all the information about allthe other vehicles. This leads to a useful subset of COP that is relevant to decision making for a particular vehicle,

e.g. a node probably needs more data about its complementary sensor vehicle than it does from a vehicle in another

remote cluster. Common Relevant Operating Picture (CROP) is a subset of COP that is then stored on each vehicle.Each vehicle communicates its information periodically during the low activity periods which is used by all the other

vehicles to synchronize their local copies of the CROP. Through this design, when a significant event occurs, decisions

are made quickly by using the latest local CROP data.

(a) Local Leader (b) Decentralized

Figure 5. (a) The Local Leader architecture allows the vehicles to be grouped according to capabilities, locale, or some other qualifier into pla-

toons with one of the platoon members responsible for assimilating data and making locally viable decisions. (b) The Decentralized architecture

requires all agents to represent themselves when deciding how to collaborate.

6 of 23


2. Decentralized implementation

The Decentralized structure, while robust to single-point-failures, does lead to potential decision conflicts if the CROPbecomes stale or inconsistent across the nodes. This may lead to multiple nodes trying to respond or no vehicle may

respond to significant events. This problem is easily solved by designing the decision rules and algorithms to be robust

against the expected levels of uncertainties and using minimal negotiations between the potentially conflicting nodes.

As discussed above, various communications topologies and architectures are driven by the mission requirements.

Though the USVs in hostile littoral warfare operations are likely to require a completely decentralized architecture,

we have designed CoMPACT to be adaptable to other architectures as well. By using well structured approaches ofArtificial Intelligence (AI), we have implemented a decision making process that can easily be migrated from individual

vehicles to a centralized location or global leader if the mission warrants it. In the following section we present our

process of arriving at the autonomy design.

IV. Artificial Intelligence and Autonomy

Since the goal of autonomy is the elimination of external human control of unmanned vehicles for prolonged peri-

ods, much of the decision-making performed by human operators is supplanted by some form and subset of ArtificialIntelligence (AI). Artificial Intelligence (AI) is the study and design of intelligent agents, where an intelligent agent is

a system that perceives its environment and takes actions which maximize its chances of success. This clearly differ-

entiates itself from systems that sometimes masquerade as being autonomous, but are in fact time-scripted waypoint-following automata which are not truly reactive to the environment. AI research uses tools and insights from many

fields, including computer science, psychology, philosophy, neuroscience, cognitive science, linguistics, operations

research, economics, control theory, probability, optimization and logic.34 AI research also overlaps with tasks such asrobotics, control systems, scheduling, data mining, logistics, speech recognition, facial recognition and many others.35

A. Intelligent Agents

All control techniques that use various AI computing approaches like neural networks, networks based on Bayesian

probability computations, fuzzy logic, machine learning, evolutionary computation and genetic algorithms, can be putinto the class of intelligent identification and control strategies. Intelligent control involves design of Intelligent Agents

(IA) which perceive and act upon the environment in pursuit of goal(s). IAs can vary in complexity from simple reflex

agents to fully adaptive learning agents based on the application requirements and constraints. However, at the core ofeach IA lies the set of decision rules or logic that it follows to perform actions in response to the perceived environment.

Figure 6 shows the basic IA concept.

Intelligent AgentEnvironment

Action

Actuator

Perception

Sensor

Decision Rules

condition−action

IF..THEN

Figure 6. Intelligent Agent Concept.

Autonomy design is then the development of decision rules that the intelligent agents use while operating in an

unstructured environment. This, in essence, defines the “behavior” of the intelligent agent. These behaviors can varyfrom being simple reactive reflex rules (bugs scattering when light is suddenly switched on, prey-predator interaction)

to highly complex adaptive learning behaviors. There are two main methods of arriving at the decision rules that define

an IA behavior:

1. Rules generated by automated learning (reinforcement or supervised, off-line or on-line); and

2. Rules generated by Domain Experts.

Choice of one of the above methods depends largely upon the specific application requirements and constraints. Insome cases, a design which incorporates both of these methods (e.g. initial rules defined by domain experts, and

7 of 23


continued learning through environmental interaction), may be better suited for a specific application.

Learning methods

Learning methods typically involve finding an input-output mapping by exploring and exploiting the environment.Various actions are attempted in response to environment inputs and the ones which receive the most reinforcement

increase their chances of being kept (Figure 7). Learning can be performed off-line using simulations against modeled

environment, or on-line while operating in real-time. Alternatively, off-line learning is followed by on-line adaptation.These methods are suitable when the environment and its inputs can be modeled with sufficient fidelity and accuracy.

Simulation coverage of expected environmental inputs becomes important in determining the performance of learned

decision rules in the presence of untested inputs. On-line learning also requires the acceptability of making mistakeswhile working in real-time, which is unacceptable in many cases.

ReinforcementActionPerception

Decision Rules

Environment

Figure 7. Reinforcement Learning.

Domain experts

Domain experts or subject matter experts (SMEs) are individuals who are both knowledgeable and extremely expe-rienced with the intended application. Domain experts are consulted when coding the decision rules for intelligent

agents. The IA should make the same (or similar) decisions that the domain experts would make while operating in the

same environment. Domain experts are also valuable in defining the concept of operation (CONOPS) for the systemwhich can guide the decision rule design process.

Since the missions involving unmanned vehicles are trying to replace dangerous or labor intensive manned mis-sions, vast amounts of human expertise available from operations like MCM and ASW can be utilized. Moreover,

real-world scenarios become very hard to simulate with sufficient fidelity, accuracy and richness that can be used by

the learning methods. Domain experts can also provide low probability but important scenarios that would be hard toreplicate in automated simulation setups.

Decision Rule Implementation

Once the method of determining the decision rules has been selected, the next choice we confront is one of the variousmethods that can be used to represent and implement these decision rules. Various possible methods are: finite state

machines (FSMs), Petri nets, decision tables, genetic algorithm (GA) strings, neural network (NN) mappings, or,

in some simple cases, even ladder logic programming. GA and NN methods are suitable for applications wherelearning methods are used and the process inherently generates these mappings. However, as discussed above, some

applications do not lend themselves easily to accurate modeling, and make the learning method results somewhat

unreliable. Domain experts method currently seems better suited for designing the decision rules for the USV missions,especially given the data available. Among other implementation methods, FSMs have been extensively used and

tested in the aerospace as well as in automotive industries for hybrid systems design. FSMs are also now starting to be

a valuable tool in fault tolerant system designs for aerospace and space applications.

8 of 23


B. Finite State Machines

A finite state machine (FSM) or automaton is a paradigm for modeling behavior composed of a finite number of states,rules for event-triggered transitions between those states and actions. In addition to their use in modeling reactive

systems presented here, finite state automata are significant in many different areas, including electrical engineering,

linguistics, computer science, philosophy, biology, mathematics, and logic. In computer science, finite state machinesare widely used in modeling of application behavior, design of hardware digital systems, software engineering, com-

pilers, network protocols, and the study of computation and languages.

Finite State Machines (FSMs) can be typically designed as a hierarchical decomposition of states/modes, tasks,subtasks, actions and action primitives. Examples at various levels may include:

• States/Modes: Search, Inspection, ISR, Build World Model, ...

• Tasks: Plan search, Detect, Localize, Reacquire, Take & send video, Interrogate, Neutralize, ...

• Subtasks: GoTo, Start, Stop, Loiter/Stationkeep, Wait, ...

• Actions: Path & Trajectory Planning, ...

• Action Primitives: Constant Velocity cruise, Accelerate, Decelerate, Turn, ...

Events guarded by specified conditions trigger transitions between states. Events and transitions may lead tospecific tasks and actions. At the lowest level, tasks and actions are implemented as suitable algorithms. For example, a

target detection event may trigger a transition for the vehicle into a pursue state, which may require an intercept transittask, which in turn requires invoking a path planning algorithm to find a safe and feasible pursuit path. FSMs can be

represented graphically using state diagrams, Figure 8, or using one of various state transitions table formats.

Using state diagrams, we designed the autonomy for the heterogeneous USV search, MCM and ASW missions.Simulation results from this design gathered tested against a low fidelity simulation are presented. Further research

requires higher fidelity world modeling as well as a more complex FSM design.

Figure 8. Finite state machine (FSM) design for a USV search and respond mission.

Due to the various threat types, it was important to include diversity in the vehicle capabilities. Vehicles equippedwith sonar, or other underwater detection methods are essential, as are vehicles carrying radar to detect surface threats.

The simulations include multiple USVs, UUVs and pairs of unmannedmine-hunting vehicles (AMVs and PMVs). The

USVs have surface searching capabilities and are faster than the other vehicles. The UUVs are included as kill vehiclesonly and are deployed from the USVs. The mine-hunting vehicles are conceptually based on Lockheed Martin’s Sea

TALON system and are deployed in pairs as one carries an active sonar array and the other carries a passive array.

Each vehicle carries specific sensors, and each sensor is simply modeled by a range and field-of-view. To advance

a threat’s status, a specific task needs to be carried out, which requires the use of a specific sensor. Thus, only certain

vehicles are capable of accomplishing certain tasks for a given threat type. Table 1 shows the breakdown of vehiclecapabilities by task and threat type.

9 of 23


Tasks

Threats Travel Observe Clear Deliver Strike Confirm

Submarine UMV UMV UMV USV UUV UMV

Mine UMV UMV UMV USV UUV UMV

Ship USV USV UMV USV UUV USV

Table 1. Vehicle capabilities chart showing which vehicles can accomplish which tasks on the included threat types. Vehicles are as follows:

UMV Unmanned Mine-hunting Vehicle pair, USV Unmanned Surface Vehicle, UUV Unmanned Underwater Vehicle. For example, a UMV

can perform the ’Clear’ task for threats of type ’Submarine’, ’Mine’, or ’Ship’.

Table 2 describes the meaning of each threat status and shows and describes the vehicle task necessary to advancethe threat’s status down the succession. The assignment algorithm uses Tables 1 and 2 together to determine vehicle

eligibility and ultimately a “score” for a given task and threat type. For example,when a ship that was not previouslyseen (i.e. hidden) becomes detected, (either through search or an intelligence alert) the required vehicle task is to

observe the threat, as seen in Table 2. From Table 1, we see that only a USV can observe a ship to classify it as a friend

or foe. Hence, the assignment to observe the ship will go the USV with the highest predicted score, disregarding allother vehicle types.

Threat Vehicle

Status Description Task Description

alert Presence of a possible threat is likely travel Vehicle must travel to within sensor rangeof the threat

detected Threat has been detected by an on-boardsensor

observeVehicle plans an observation path aroundthe threat for positive classification

posIDThreat has been identified as either friendor foe

clearIf threat is foe, path is planned to check andclear the immediate area of other possiblethreats

cleared Immediate surroundings are clear of otherthreats

deliver Path is planned to deliver the kill vehiclesto the threat area

marked Kill vehicles are in place strike Kill vehicles are given paths to interceptthreat

hit Threat has been hit by kill vehicles confirm Plan a path to observe the threat and con-firm that it has been killed

killed Threat has been destroyed and/or disabled none

Table 2. Threat status descriptions and associated vehicle tasks with descriptions

C. Dynamic Mission Planning and Task Assignment in the Context of Multiple USVs

We used the vehicle assignment algorithm based on a performance criterion and constraints arising from the capabilitymatrix so that the related cost and constraints depend on the following for each individual USV: (i) sensor suite, (ii)

sensor characteristics, (iii) maximum acceleration, turn rate and velocity, (iv) available fuel, and (v) current distance

from the target.

The key element in this cost are the paths generated by the Rapidly-exploring Random Trees (RRT) algorithm that

enable rapid calculation of the distances between the vehicle and the target, and allow quick recalculations in the caseof path re-planning for pop-up threat avoidance. The cost also takes timing constraints into account and, essentially,

represents a time-parameterized greedy algorithm.

However, it is also clear that other assignment algorithms arising in the context of the multi-agent task allocation(MATA) problem are fully applicable to coordinated mission management and autonomy for multiple USVs. In all

cases these algorithms can be implemented as Dynamic Task Allocation (DTA) strategies. This is because, regardlessof the technique chosen to address the allocation problem, the assignment is recalculated whenever there is a change

in the system state or environment. Hence this is an event-driven DTA where, for instance, events include ”task

accomplished”, “presence of new targets”, “presence of pop-up threats”, etc. It should be noted that the DTA is closelyaffected by the sensor capabilities and communications architecture used.

10 of 23


V. Levels of Autonomy in Autonomous Control of Multiple USVs

A primary objective of the project was to reduce the operator workload during multi-USV missions in complexlittoral environments. One of the key issues in this context is a suitable framework within which the autonomy levels

can be defined in a meaningful way. The goal of studying autonomy levels is to arrive at a thorough understanding

of the operator role at each level, and to correlate the “amount” of autonomy with the operator workload. This hasbeen set within the framework of CoMPACT system allowing, at the highest level, substantially reduced workload for

individual operators, along with the reduction in number of operators needed to handle a large fleet of autonomous

USVs.

As an examplemission to define autonomy levels, we have chosen the ship inspectionmission carried out by a single

USV. The objective is to use the USV, starting from some initial position at initial time, to inspect a stationary ship bytaking a video, and return to the starting point. The video stream is processed to carry out Automatic Target Recognition

(ATR). Several levels of autonomy have been defined in this context, from fully manual to fully autonomous.

Our first step was to define autonomy levels in the context of this specific mission. The resulting levels are basedon the basic tasks during a ship inspection mission, but can easily be abstracted to other autonomous vehicle missions.

These include: (i) USV trajectory tracking control; (ii) Inspection data processing; and (iii) Dynamic reactive re-

planning.

Level 1: Tele-Operation. In this case all the tasks are carried out manually (trajectory following, data processing,re-planning). This level is characterized by the highest operator load, and no autonomous control.

Level 2: Assisted Tele-Operation. In this case the CoMPACT system generates waypoints corresponding to the mis-sion path, and also suggests optimal or sub-optimal trajectories displayed on the mission GUI. The operator manually

steers the USV through the waypoints trying to also track the trajectory. All other tasks are also carried out by the

operator.

Level 3: Semi-Autonomous Operation. CoMPACT generates waypoints based on the ship location, as well as the

trajectory and outer-loop control & guidance. Reactive task or mission re-planning is carried out by the operator, while

ATR is carried out at the command station

Level 4: Autonomous Operation. Besides the tasks from Level 3, the CoMPACT also carries out reactive re-planning

of tasks or mission segments. ATR is still carried out at the command station

Level 5: Fully Autonomous Mission. Besides the tasks from Level 4, CoMPACT also carries out ATR and declares

“Friend” or “Foe”.

Relationships between different autonomy levels are illustrated in Figure 9.

In more general cases, different autonomy levels are applicable to a variety of missions carried out with a single

UAV. These levels are consistent with ASTM Standard F2541.29298-1: Guide for Autonomy and Control of UnmannedMaritime Vehicles.

VI. Autonomous Search using Multiple USVs

The idea of a cooperative reconfigurable search is to dynamically allocate resources to guarantee coverage of a

specified region in the presence of a changing environment and varying available resources. Under this project, theavailable resources are unmanned agents, and the littoral environment changes due to pop-up obstacles and threats.

Any vehicle participating in the search can become unavailable at any time due to a higher priority task or asystem malfunction. It is assumed that a vehicle becoming unavailable will make its status known to the team either

through communications, in the case of a higher priority task, or through communication silence in the case of a

system malfunction. The remaining vehicles then cooperate to reconfigure their search patterns to provide coverage ofthe search region while minimizing the number of redundant search cells. Cooperative reconfigurable search-related

simulation results are included below.

Figure 10(a) shows the initial layout of the simulation. All threats are stationary and the search area is pre-defined.All vehicles will initially participate in the search. Since all agents initially participate in the search, the area is divided

into 5 equal cones and each agent is assigned to search a cone and is given a path that accommodates this assignment.

These paths are shown in Figure 10(b). In Figure 11(a) we see that the vehicles’ paths initially cover the search areasuch that if no re-plans are necessary, coverage of the search region is guaranteed. In Figure 11(b) we see that one of

the vehicles is no longer available to participate in the search. As seen in the Simulation Messages window on the rightside of the figure, Vehicle 3 is not available. This simulates a variety of situations, such as when a vehicle becomes

disabled and can no longer function as a member of the team. Becoming unavailable in this manner can also simulate

a vehicle being pulled from the search to service a threat outside of the search area.

11 of 23


(a) Levels of Autonomy

(b) Levels of Autonomy Details

Figure 9. The levels of autonomy for ship inspection.

12 of 23


0 5 10 15 20 25 300

5

10

15

20

25

30

East

No

rth

1

2

Time: 0

Simulation Messages:

Vehicle Status:

Threat Status:

(a)

Time: 0

Simulation Messages: T= 0.0: Replan search − Vehicle Status Change

Vehicle Status:

Threat Status:

0 5 10 15 20 25 300

5

10

15

20

25

30

East

No

rth

1

2

(b)

Figure 10. (a) The initial layout of the search area, USV team, and threats. (b) The search region is divided among available agents and

initial assignments are made.

Continuing, we see in Figure 12(a) paths are successfully re-planned for the remaining agents that cover the un-

searched portion of the search area. Notice that the region searched by Vehicle 3 prior to it becoming unavailable wasrecorded and will not be searched again. Figure 12(b) shows that Vehicle 3 is now available to join the search. This

triggers another cooperative reconfiguration of the search pattern. Figure 13(a) shows the team of agents cooperatingto reconfigure the search patterns and include a new vehicle in the search. Figure 13(b) shows the search is complete,

leaving all participating agents along an arc awaiting a new assignment or ready for easy recovery by an LCS.

Time: 0.0

Simulation Messages: T= 0.0: Replan search − Vehicle Status Change

Vehicle Status: 1 USV: search 2 USV: search 3 USV: search 4 USV: search 5 USV: search

Threat Status:

1−

hid

de

n

2−

hid

de

n

4 6 8 10 12 14 16

10

12

14

16

18

20

22

East

No

rth

1

(a)

Time: 402.5

Simulation Messages: T= 0.0: Replan search − Vehicle Status Change T=402.5: Vehicle 3 is not available

Vehicle Status: 1 USV: search 2 USV: search 3 USV: disabled 4 USV: search 5 USV: search

Threat Status:

1−

hid

de

n

2−

hid

de

n

4 6 8 10 12 14 16

10

12

14

16

18

20

22

East

No

rth

1

(b)

Figure 11. (a) The agents’ paths are designed to cover the region of interest. (b) One of the vehicles is no longer available to participate in

the search.

13 of 23


Time: 405.0

Simulation Messages: T= 0.0: Replan search − Vehicle Status Change T=402.5: Vehicle 3 is not available T=405.0: Replan search − Vehicle Status Change

Vehicle Status: 1 USV: search 2 USV: search 3 USV: disabled 4 USV: search 5 USV: search

Threat Status:1−

hid

den

2−

hid

den

4 6 8 10 12 14 16

10

12

14

16

18

20

22

East

No

rth

1

(a)

Time: 902.5

Simulation Messages: T= 0.0: Replan search − Vehicle Status Change T=402.5: Vehicle 3 is not available T=405.0: Replan search − Vehicle Status Change T=902.5: Vehicle 3 is available

Vehicle Status: 1 USV: search 2 USV: search 3 USV: loiter 4 USV: search 5 USV: search

Threat Status:

1−

hid

den

2−

hid

den

4 6 8 10 12 14 16

10

12

14

16

18

20

22

East

No

rth

1

(b)

Figure 12. (a) The remaining available agents cooperatively reconfigure their search paths to cover the region left open byVehicle 3 becoming

unavailable. (b) Vehicle 3 has become available to rejoin the search.

Time: 905.0

Simulation Messages: T= 0.0: Replan search − Vehicle Status Change T=402.5: Vehicle 3 is not available T=405.0: Replan search − Vehicle Status Change T=902.5: Vehicle 3 is available T=905.0: Replan search − Vehicle Status Change

Vehicle Status: 1 USV: search 2 USV: search 3 USV: search 4 USV: search 5 USV: search

Threat Status:

1−

hid

de

n

2−

hid

de

n

4 6 8 10 12 14 16

10

12

14

16

18

20

22

East

No

rth

1

(a)

Time: 1955.0

Simulation Messages: T=402.5: Vehicle 3 is not available T=405.0: Replan search − Vehicle Status Change T=902.5: Vehicle 3 is available T=905.0: Replan search − Vehicle Status Change T=1655.0: Task complete for veh 1 T=1657.5: Replan search − Vehicle Status Change T=1742.5: Task complete for veh 2 T=1742.5: Task complete for veh 5 T=1745.0: Replan search − Vehicle Status Change T=1765.0: Task complete for veh 3 T=1767.5: Replan search − Vehicle Status Change T=1955.0: Task complete for veh 4Vehicle Status: 1 USV: loiter 2 USV: loiter 3 USV: loiter 4 USV: loiter 5 USV: loiter

Threat Status:

1−

hid

de

n

2−

hid

de

n

4 6 8 10 12 14 16

10

12

14

16

18

20

22

East

No

rth

1

(b)

Figure 13. (a) The remaining search area is cooperatively reconfigured among the available vehicles. (b) The search is complete.

14 of 23


VII. Fully Autonomous MCM & ASW

A mission scenario used for this work consists of the following tasks: (i) Predefined search, (ii) Detection of asuspicious vessel, (iii) Assignment of one or more USVs from the team to inspect the vessel, (iv) Communication of

the results of inspection to the control center, (v) Re-tasking of the remaining USVs to continue the search, (vi) Return

of the USV that has carried out the inspection to the search team, and (vii) USV team re-tasking to continue the searchwith an additional vehicle.

0 5 10 15 20 25 300

5

10

15

20

25

30

East

No

rth

Time: 0.0


Vehicle Status: 1 LCS: loiter 2 UAV: loiter 3 AMV: loiter 4 AMV: loiter 5 PMV: loiter 6 PMV: loiter 7 USV: loiter 8 USV: loiter 9 UUV: loiter 10 UUV: loiter 11 UUV: loiter 12 UUV: loiter

Threat Status:

hid

de

n

hid

de

n

hid

de

n

hid

de

n

hid

de

n

hid

de

n

hid

de

n

hid

de

n

USVs

UAV

Passive Mine−hunting Vehicles

Active Mine−hunting Vehicles

LCS

Mine threats Submarine threat

Ship threats

Known obstacles

Figure 14. A littoral region was chosen as the theater of operations for our mission scenario. One area of the region was chosen for a

cooperative search mission and multiple threats are scattered throughout the region. The mine threats are initially at unknown locations.

Figure 14 shows the mission scenario’s layout and theater of operation. As seen, the littoral environment canharbor many different threat types, such as submarines, mines and ships. A heterogeneous team of unmanned vehicles

is deployed from a Littoral Combat Ship (LCS). The team consists of multiple unmanned surface vehicles (USVs),

active and passive pairs of unmanned mine-hunting vehicles (AMVs, PMVs) and unmanned underwater kill vehicles(UUVs), the latter being deployed from the USVs. The unmanned vehicle team immediately begins a cooperative area

search, scanning for both surface and subsurface threats. When a threat is detected, or when intelligence alerts the team

of a possible threat, a “re-plan” is triggered amongst all team members. Associated with this re-plan is key informationabout the possible threat, such as its probable type and location. A specific vehicle task is then derived and each vehicle

capable of performing this task on the threat submits an associated “score”. The vehicle, or vehicle pair, depending on

the task, with the highest “score” is chosen to carry out the required task. Those vehicles not assigned to service thethreat remain assigned to the cooperative area search, which must be reconfigured due to the now fewer available team

members.

The vehicles servicing the threat rejoin the cooperative area search upon assignment completion and the search is

reconfigured to accommodate the additional resources. This process of dynamically reassigning vehicles to and from

the area search to observe, identify, strike, confirm and otherwise service various potential threats continues until thecooperative area search is complete. Note that the autonomy involved in the MCM/ASW activities is designed to be

resilient amid sparse obstacle fields. Next we present results on maneuvering amid dense obstacle fields using various

levels of autonomy, and describe the results of our mission scenario simulation.

15 of 23


The simulation has the following capabilities: (i) Team-to-target assignment algorithms, where teams of two or

more vehicles are assigned to the target based on a capability matrix and cost minimization; (ii) Simultaneous arrival

of two or more vehicles to the target area with prespecified headings; (iii) Pop-up threat avoidance with velocityadjustment to guarantee simultaneous arrival; and (iv) Vehicle loitering and waiting for the next assignment.

The simulation also includes: typical vehicle velocities; an arbitrary theater of operation size; 1 LCS to deploy theteam of unmanned vehicles ; 1 UAV to provide communication between the team and the LCS; 2 UMV pairs (1 UMV

pair = 1 AMV + 1 PMV) to service the subsurface-related tasks and threats; 2 USVs to service the surface-related tasks

and threats; 3 UUVs on board each USV (6 total) to strike unfriendly threats; 1 submarine threat; 1 mine threat; and 1ship threat. As mentioned previously, each USV has a simplified sensor model describing its range, performance and

field of view. The following results present the results of the USVs that were pulled from the simulation of cooperative

area search. This simulation is focused on task generation and the assignment of vehicles to pop-up threats to advancethe status of each detected threat to either friend or killed. The assignment algorithm is built around a “re-plan” flag

which indicates one of the following three states:

1. Keep all current assignments, no re-plan needed

2. Obstacle detected in vehicle i’s path, re-plan around this assignment. In other words, pick the best vehicle (not

necessarily vehicle i) to accomplish the task vehicle i was assigned to before it detected an obstacle in its path.

3. The status of one or more threats has changed, issue a global re-plan. A global re-plan assigns the best vehicleto each threat in need of a task. Often, this results in vehicles being reassigned to the same task they previously

had. However, this method is kept for generality.

Multiple snapshots from the simulation described above are included.

Figure 15(a) shows the initial layout of the simulation. All threats are stationary, as is the LCS. The UAV launchesfrom the LCS and loiters around the midpoint between the farthest unmanned vehicle and the LCS. All USVs (orange)

and UMV pairs (red = AMV, green = PMV) not assigned to MCM or ASW tasks are assigned to cooperative area

search.

Notice in Figure 15(b) in the Simulation Messages window on the right side of the Figure, a message indicates that

“Intel alerts of Threat 3” and “Intel alerts of Threat 1”. This message window is updated at every time step with themost recent messages at the bottom of the stack. Also seen in Figure 15(b), assignments are made to travel to Threat 1

and Threat 3 so they can be detected with vehicle sensors.

Since all agents initially participate in the search, the area is divided into 5 equal cones and each agent is assignedto search a cone and is given a path that accommodates this assignment. These paths are shown in Figure 15(b).

0 5 10 15 20 25 300

5

10

15

20

25

30

East

No

rth

1

2

3

Time: 0.0


Vehicle Status: 1 LCS: loiter 2 UAV: loiter 3 AMV: loiter 4 AMV: loiter 5 PMV: loiter 6 PMV: loiter 7 USV: loiter 8 USV: loiter

Threat Status:

1−

hid

de

n

2−

hid

de

n

3−

hid

de

n

(a)

Time: 422.5

Simulation Messages: T=72.5: Intel alerts of Threat 3 T=72.5: REPLAN, Threat 3 > alert T=110.0: Intel alerts of Threat 1 T=110.0: REPLAN, Threat 1 > alert T=422.5: Obstacle in path, veh 8 replan

Vehicle Status: 1 LCS: loiter 2 UAV: loiter 3 AMV: loiter 4 AMV: travel on tgt 1, eta = 1406.2 5 PMV: loiter 6 PMV: travel on tgt 1, eta = 1396.2 7 USV: loiter 8 USV: travel on tgt 3, eta = 704.9

Threat Status:

1−

ale

rt

2−

hid

de

n

3−

ale

rt

0 5 10 15 20 25 300

5

10

15

20

25

30

East

No

rth

1

2

3

(b)

Figure 15. (a) Initial layout of the LCS, UAV, unmanned vehicle team and threats. (b) The vehicle team is alerted of threats #1 and #3 and

assignments are made.

In Figure 16(a) we see that the USV (orange) has detected an obstacle in its path. In this case, the re-plan flag

was set to 2 to initiate a re-plan for this assignment only. As seen by the adjusted path (orange), the same vehicle was

chosen again to travel to Threat 3.

In Figure 16(b) we see that a UMV pair (red + green) have detected a possible threat. This is indicated in the

Simulation Messages window as well as by the outline of the gray threat turning yellow. Continuing, we see inFigure 17(a) that the same UMV pair that detected the possible threat (Threat 2) has been assigned to observe the threat

16 of 23


Time: 452.5

Simulation Messages: T=72.5: Intel alerts of Threat 3 T=72.5: REPLAN, Threat 3 > alert T=110.0: Intel alerts of Threat 1 T=110.0: REPLAN, Threat 1 > alert T=422.5: Obstacle in path, veh 8 replan T=450.0: Obstacle in path, veh 8 replan T=452.5: Obstacle in path, veh 8 replan


Threat Status:

1−

ale

rt

2−

hid

den

3−

ale

rt

2 4 6 8 10 12 14 16 18 20 22

2

4

6

8

10

12

14

16

18

20

22

East

No

rth

1

2

3

(a)

Time: 622.5

Simulation Messages: T=72.5: Intel alerts of Threat 3 T=72.5: REPLAN, Threat 3 > alert T=110.0: Intel alerts of Threat 1 T=110.0: REPLAN, Threat 1 > alert T=422.5: Obstacle in path, veh 8 replan T=450.0: Obstacle in path, veh 8 replan T=452.5: Obstacle in path, veh 8 replan T=622.5: Veh 4 detects Threat 2 T=622.5: REPLAN, Threat 2 > detected


Threat Status:

1−

ale

rt

2−

dete

cte

d

3−

ale

rt

2 4 6 8 10 12 14 16 18 20 22

2

4

6

8

10

12

14

16

18

20

22

East

No

rth

1

2

3

(b)

Figure 16. (a) The USV (orange) detects an obstacle en-route and reacts, planning a new, safe path. (b) A UMV pair detect a mine en-route.

Time: 672.5

Simulation Messages: T=72.5: Intel alerts of Threat 3 T=72.5: REPLAN, Threat 3 > alert T=110.0: Intel alerts of Threat 1 T=110.0: REPLAN, Threat 1 > alert T=422.5: Obstacle in path, veh 8 replan T=450.0: Obstacle in path, veh 8 replan T=452.5: Obstacle in path, veh 8 replan T=622.5: Veh 4 detects Threat 2 T=622.5: REPLAN, Threat 2 > detected T=672.5: Veh 8 detects Threat 3 T=672.5: REPLAN, Threat 3 > detected

Vehicle Status: 1 LCS: loiter 2 UAV: loiter 3 AMV: travel on tgt 1, eta = 1972.4 4 AMV: observe on tgt 2, eta = 1095.8 5 PMV: travel on tgt 1, eta = 1962.4 6 PMV: observe on tgt 2, eta = 1085.8 7 USV: loiter 8 USV: travel on tgt 3, eta = 720.7

Threat Status:

1−

ale

rt

2−

de

tecte

d

3−

de

tecte

d

2 4 6 8 10 12 14 16 18 20 22

2

4

6

8

10

12

14

16

18

20

22

East

No

rth

1

2

3

(a)

Time: 1040.0

Simulation Messages: T=72.5: REPLAN, Threat 3 > alert T=110.0: Intel alerts of Threat 1 T=110.0: REPLAN, Threat 1 > alert T=422.5: Obstacle in path, veh 8 replan T=450.0: Obstacle in path, veh 8 replan T=452.5: Obstacle in path, veh 8 replan T=622.5: Veh 4 detects Threat 2 T=622.5: REPLAN, Threat 2 > detected T=672.5: Veh 8 detects Threat 3 T=672.5: REPLAN, Threat 3 > detected T=837.5: REPLAN, Threat 3 > posID T=1040.0: REPLAN, Threat 2 > posIDVehicle Status: 1 LCS: loiter 2 UAV: loiter 3 AMV: clear on tgt 3, eta = 3014.7 4 AMV: observe on tgt 2, eta = 1096.5 5 PMV: clear on tgt 3, eta = 3004.7 6 PMV: observe on tgt 2, eta = 1086.5 7 USV: loiter 8 USV: loiter

Threat Status:

1−

ale

rt

2−

po

sID

3−

po

sID2 4 6 8 10 12 14 16 18 20 22

2

4

6

8

10

12

14

16

18

20

22

East

No

rth

1

2

3

(b)

Figure 17. (a) One UMV pair observes the newly detected mine, while the other is assigned to travel to threat #1. The USV has detected

threat #3, and planned an observe path around it. (b) The USV has identified threat #3 as unfriendly and called a UMV pair to clear the

area of further threats. Also, a UMV pair has identified threat #2 as a mine and are beginning to clear the area. Progress on threat #1 is

halted due to lack of resources.

and attempt to classify it. Also shown in Figure 17(a) the USV has detected Threat 3 and has now been assigned toobserve it. These assignments match those organized in Table 1 and Table 2.

Figure 17(b) shows some interesting features of the assignment algorithm. All threats are in need of some task thatrequires the use of a UMV pair. Threat 1 needs to be detected, but since it is a submarine, it can only be detected by a

UMV pair. Threat 2 has been positively identified and needs the area surrounding it cleared in preparation for striking

it. To clear an area requires checking for mines and other threats, this is always carried out with a UMV pair, regardlessof threat type. Threat 3 has also been positively identified and needs its surrounding area cleared. However, there are

only 2 UMV pairs, so the assignment algorithm assigns the two pairs to the tasks with the highest score (i.e. lowest

cost). The threat with the lowest score (Threat 1) is left to wait for available resources, while Threat 2 and Threat 3have their surroundings cleared by the available UMV pairs.

Figure 18(a) shows that the area surrounding Threat 2 has been successfully cleared and this UMV pair has beenassigned to clear Threat 1, who has been waiting. At the same time the UMV pair is reassigned to Threat 1, a USV is

assigned to Threat 2 to carry out the deliver task required to advance the status of Threat 2 from cleared to marked.

After the UUVs have been delivered to the area, as seen in Figure 18(b), they are assigned a strike task to attemptto kill the threat. It is seen that the two UUVs hit their target, but it will need to be confirmed that the target is killed.

The UMV pairs continue their previous assignments. When one pair is finished, they will become available to confirm

the hit on Threat 2.

In Figure 19(a) we see that the UMV pair traveling to Threat 1 have detected it and have been reassigned to now

observe it. Note that all vehicles in this simulation are NOT to drawn to scale against the littoral environment.

17 of 23


In Figure 19(b), we see that Threat 1 has been positively identified and the UMV pair previously assigned to observe

it has now been assigned to clear the area surrounding it, in preparation for a strike. The other UMV pair has finished

clearing the area around Threat 3 and have been reassigned to confirm the hit on Threat 2 made earlier. Meanwhile aUSV has been assigned to deliver UUVs to the area near Threat 3, which has just been cleared.

Time: 2315.0

Simulation Messages: T=110.0: REPLAN, Threat 1 > alert T=422.5: Obstacle in path, veh 8 replan T=450.0: Obstacle in path, veh 8 replan T=452.5: Obstacle in path, veh 8 replan T=622.5: Veh 4 detects Threat 2 T=622.5: REPLAN, Threat 2 > detected T=672.5: Veh 8 detects Threat 3 T=672.5: REPLAN, Threat 3 > detected T=837.5: REPLAN, Threat 3 > posID T=1040.0: REPLAN, Threat 2 > posID T=2315.0: Task complete for veh 6 T=2315.0: REPLAN, Threat 2 > clearedVehicle Status: 1 LCS: loiter 2 UAV: loiter 3 AMV: clear on tgt 3, eta = 3015.6 4 AMV: loiter 5 PMV: clear on tgt 3, eta = 3005.6 6 PMV: loiter 7 USV: loiter 8 USV: loiter

Threat Status:

1−

ale

rt

2−

cle

are

d

3−

posID2 4 6 8 10 12 14 16 18 20 22

2

4

6

8

10

12

14

16

18

20

22

East

No

rth

1

2

3

(a)

Time: 2752.5

Simulation Messages: T=1040.0: REPLAN, Threat 2 > posID T=2315.0: Task complete for veh 6 T=2315.0: REPLAN, Threat 2 > cleared T=2710.0: Task complete for veh 7 UUV 1 deployed to strike Threat 0 UUV 2 deployed to strike Threat 0 T=2710.0: REPLAN, Threat 2 > marked T=2752.5: Task complete for veh 9 Successful strike, Threat 2 hit T=2752.5: Task complete for veh 10 Successful strike, Threat 2 hit T=2752.5: REPLAN, Threat 2 > hitVehicle Status: 1 LCS: loiter 2 UAV: loiter 3 AMV: clear on tgt 3, eta = 3030.8 4 AMV: travel on tgt 1, eta = 3059.4 5 PMV: clear on tgt 3, eta = 3020.8 6 PMV: travel on tgt 1, eta = 3049.4 7 USV: loiter 8 USV: loiter

Threat Status:

1−

ale

rt

2−

hit

3−

posID8 10 12 14 16 18 20 22

8

10

12

14

16

18

20

22

East

No

rth

1

2

3

(b)

Figure 18. (a) The area around threat #2 is successfully cleared and this UMV pair move up to observe threat #1. A USV is called in to

deliver the kill UUVs to threat #2 while a UMV pair continues to clear the area surrounding threat #3. (b) Two kill UUVs were delivered to

threat #2, which was hit. UMV pairs continue to travel to threat #1 and clear the area around threat #3.

Time: 2917.5

Simulation Messages: T=2315.0: REPLAN, Threat 2 > cleared T=2710.0: Task complete for veh 7 UUV 1 deployed to strike Threat 0 UUV 2 deployed to strike Threat 0 T=2710.0: REPLAN, Threat 2 > marked T=2752.5: Task complete for veh 9 Successful strike, Threat 2 hit T=2752.5: Task complete for veh 10 Successful strike, Threat 2 hit T=2752.5: REPLAN, Threat 2 > hit T=2917.5: Veh 6 detects Threat 1 T=2917.5: REPLAN, Threat 1 > detectedVehicle Status: 1 LCS: loiter 2 UAV: loiter 3 AMV: clear on tgt 3, eta = 3032.2 4 AMV: travel on tgt 1, eta = 3060.8 5 PMV: clear on tgt 3, eta = 3022.2 6 PMV: travel on tgt 1, eta = 3050.8 7 USV: loiter 8 USV: loiter

Threat Status:

1−

de

tecte

d

2−

hit

3−

po

sID8 10 12 14 16 18 20 22

8

10

12

14

16

18

20

22

East

No

rth

1

2

3

(a)

Time: 3492.5

Simulation Messages: Successful strike, Threat 2 hit T=2752.5: REPLAN, Threat 2 > hit T=2917.5: Veh 6 detects Threat 1 T=2917.5: REPLAN, Threat 1 > detected T=3027.5: Task complete for veh 5 T=3027.5: REPLAN, Threat 3 > cleared T=3412.5: Task complete for veh 6 T=3427.5: REPLAN, Threat 1 > posID T=3492.5: Task complete for veh 7 UUV 1 deployed to strike Threat 0 UUV 2 deployed to strike Threat 0 T=3492.5: REPLAN, Threat 3 > markedVehicle Status: 1 LCS: loiter 2 UAV: loiter 3 AMV: confirm on tgt 2, eta = 4757.5 4 AMV: clear on tgt 1, eta = 4963.2 5 PMV: clear on tgt 1, eta = 4953.2 6 PMV: confirm on tgt 2, eta = 4747.5 7 USV: deliver on tgt 3, eta = 3490.6 8 USV: loiter

Threat Status:

1−

po

sID

2−

hit

3−

ma

rke

d8 10 12 14 16 18 20 22

8

10

12

14

16

18

20

22

East

No

rth

1

2

3

(b)

Figure 19. (a) A UMV pair have detected threat #1 and have begun to observe in order to identify it. Another UMV pair continues to clear

the area around threat #3. (b) Threat #3 is cleared, and the UMV pair are assigned to confirm the hit on threat #2. Also, a USV is assigned

to deliver UUV kill vehicles to threat #3 while a UMV pair has been assigned to clear the area surrounding threat #1 as it was just positively

identified as unfriendly.

In Figure 20(b) it is shown that Threat 3 has been hit and the USV that delivered the UUVs has been assigned to

confirm the hit and pronounce the threat as killed. The UMV pairs are continuing with their previous assignments,

namely to clear the area around Threat 1 and to confirm the hit on Threat 2.

Both Threat 2 and Threat 3 were pronounced killed after the hits were confirmed. The area surrounding Threat 1

has been cleared and a USV has been assigned to deliver kill UUVs nearby, as noted in the Messages of Figure 21(a).Also in that Figure, the USV delivers UUVs to the front door of Threat 1. The UMV pairs ready for reassignment since

there are no threats in need of tasks that require the use of a UMV pair. These vehicles are sent back to assist in the

cooperative area search.

In Figure 21(b) we see that the UUVs have hit Threat 1, and the UMV pair nearest have been assigned to confirm

the hit and declare the threat killed. The USV has successfully confirmed that Threat 1 is killed. All vehicles are now

available for participating in the cooperative area search as a task with low priority. Other tasks involving potentialthreats will take precedence over the search, requiring the search to reconfigure as often as the shifting resources

requires.

18 of 23


Time: 3492.5

Simulation Messages: Successful strike, Threat 2 hit T=2752.5: REPLAN, Threat 2 > hit T=2917.5: Veh 6 detects Threat 1 T=2917.5: REPLAN, Threat 1 > detected T=3027.5: Task complete for veh 5 T=3027.5: REPLAN, Threat 3 > cleared T=3412.5: Task complete for veh 6 T=3427.5: REPLAN, Threat 1 > posID T=3492.5: Task complete for veh 7 UUV 1 deployed to strike Threat 0 UUV 2 deployed to strike Threat 0 T=3492.5: REPLAN, Threat 3 > markedVehicle Status: 1 LCS: loiter 2 UAV: loiter 3 AMV: confirm on tgt 2, eta = 4757.5 4 AMV: clear on tgt 1, eta = 4963.2 5 PMV: clear on tgt 1, eta = 4953.2 6 PMV: confirm on tgt 2, eta = 4747.5 7 USV: deliver on tgt 3, eta = 3490.6 8 USV: loiter

Threat Status:

1−

posID

2−

hit

3−

mark

ed

17.5 18 18.5 19 19.5 20

10

10.5

11

11.5

12

12.5

East

No

rth

3

(a)

Time: 3535.0

Simulation Messages: T=3027.5: REPLAN, Threat 3 > cleared T=3412.5: Task complete for veh 6 T=3427.5: REPLAN, Threat 1 > posID T=3492.5: Task complete for veh 7 UUV 1 deployed to strike Threat 0 UUV 2 deployed to strike Threat 0 T=3492.5: REPLAN, Threat 3 > marked T=3535.0: Task complete for veh 11 Successful strike, Threat 3 hit T=3535.0: Task complete for veh 12 Successful strike, Threat 3 hit T=3535.0: REPLAN, Threat 3 > hitVehicle Status: 1 LCS: loiter 2 UAV: loiter 3 AMV: confirm on tgt 2, eta = 4428.9 4 AMV: clear on tgt 1, eta = 4656.4 5 PMV: confirm on tgt 2, eta = 4418.9 6 PMV: clear on tgt 1, eta = 4646.4 7 USV: loiter 8 USV: loiter

Threat Status:

1−

posID

2−

hit

3−

hit

10 12 14 16 18 20 22

10

12

14

16

18

20

22

East

No

rth

1

2

3

(b)

Figure 20. (a) Figure 20(a) shows a nice closeup of the UUV kill vehicles being deployed from the USV and setting up to strike Threat 3.

(b) The UUVs have hit threat #3 and a USV has been assigned to confirm the hit. Meanwhile, UMV pairs continue to clear threat #1 and

confirm the hit on threat #2.

Time: 4942.5

Simulation Messages: Successful strike, Threat 3 hit T=3535.0: REPLAN, Threat 3 > hit T=3800.0: Task complete for veh 7 T=3800.0: REPLAN, Threat 3 > killed T=4420.0: Task complete for veh 5 T=4420.0: REPLAN, Threat 2 > killed T=4657.5: Task complete for veh 6 T=4657.5: REPLAN, Threat 1 > cleared T=4940.0: Task complete for veh 7 UUV 1 deployed to strike Threat 0 UUV 2 deployed to strike Threat 0 T=4940.0: REPLAN, Threat 1 > markedVehicle Status: 1 LCS: loiter 2 UAV: loiter 3 AMV: loiter 4 AMV: loiter 5 PMV: loiter 6 PMV: loiter 7 USV: loiter 8 USV: loiter

Threat Status:

1−

ma

rke

d

2−

kill

ed

3−

kill

ed

10 12 14 16 18 20 22

10

12

14

16

18

20

22

East

No

rth

1

2

3

(a)

Time: 4982.5

Simulation Messages: T=4420.0: REPLAN, Threat 2 > killed T=4657.5: Task complete for veh 6 T=4657.5: REPLAN, Threat 1 > cleared T=4940.0: Task complete for veh 7 UUV 1 deployed to strike Threat 0 UUV 2 deployed to strike Threat 0 T=4940.0: REPLAN, Threat 1 > marked T=4982.5: Task complete for veh 13 Successful strike, Threat 1 hit T=4982.5: Task complete for veh 14 Successful strike, Threat 1 hit T=4982.5: REPLAN, Threat 1 > hitVehicle Status: 1 LCS: loiter 2 UAV: loiter 3 AMV: loiter 4 AMV: loiter 5 PMV: loiter 6 PMV: loiter 7 USV: loiter 8 USV: loiter

Threat Status:

1−

hit

2−

kill

ed

3−

kill

ed

10 12 14 16 18 20 22

10

12

14

16

18

20

22

East

No

rth

1

2

3

(b)

Figure 21. (a) A USV has delivered the UUVs to threat #1 and they are ready to strike. (b) Threat #1 has been hit, and a UMV pair is

assigned to confirm the hit. Once the Threat is confirmed killed, the simulation ends.

19 of 23


VIII. Autonomy Levels for Target Intercept and Inspection

One of the main features of USVs is their ability to autonomously intercept and inspect potential threats andsuspicious vessels. Achieving accurate intercept is essential to the success of the USV missions. Often, allowing the

USV greater levels of autonomy over path planning, trajectory generation and control result in more efficient navigation

through dense obstacle fields to the potential threat.

With the terminal trajectory concepts developed, and the levels of autonomy designed for CoMPACT in Chapter

2, the task of integrating the two results in an efficient ship inspection strategy that has a great potential for reducing

operator workload. The technology is demonstrated by developing a Graphical User Interface (GUI) so that an operatormay assume the role of path planner, controller, trajectory generator, and motion re-planner, according to the defined

levels of autonomy. The GUI design allows the operator to control heading and velocity through keyboard inputs,subject to the USV’s dynamic constraints. The GUI accommodates multiple USVs which can be commanded to travel

in formation. An illustration of the GUI is shown in Figure 22.

Figure 22. The levels of Autonomy were implemented in a Graphical User Interface to attribute varying amounts of autonomy to the USVs

A. Levels of Autonomy Simulations for Single USV

As an exercise to demonstrate the levels of autonomy integrated in CoMPACT, the objective of the tests were to

maneuver through an obstacle field to reach a target moving in open water, then inspect the moving target.

For a single USV, the autonomy levels varied from the operator being responsible for all decision, control, and

planning, to the operator chasing a virtual “carrot,” to turning all control and planning to CoMPACT. Through these

simulations, it was observed that unless the obstacle field was very sparse, avoiding the obstacles was very tedious.Furthermore, performing ship inspection is very difficult without autonomous control. Results from the autonomous

ship inspection using a single USV are presented in Figures 23(a) through 24(b) .

B. Levels of Autonomy Simulations for Two USVs

When the operator is tasked with controlling two USVs, monitoring both vehicle’s progress individually proves a

monumental problem. However, when the operator is guiding the USVs manually, if they are virtually linked together

through a task to travel in formation, a single operator is still able to manage both vessels. Enabling the vessels tomaintain autonomous formations helps ensure successful traversal through the obstacle field, but turning the CoMPACT

system to fully autonomous further increases the likelihood of efficient inspection of suspicious moving targets.

20 of 23


(a) (b)

Figure 23. (a) Level 1 Autonomy: The USV is driven solely by operator control with no path-planning assistance. Amid obstacle fields,

avoiding collisions requires rapt attention. (b) Level 2 Autonomy: CoMPACT evaluates the obstacle field and develops a recommended

path for the USV to follow. The moving red star represents the desired position for following the trajectory through a successful inspection.

(a) (b)

Figure 24. (a) Level 2 Autonomy with Replanning: Even when a recommendation is given, if the operator decides to take a different

course than the recommended trajectory, CoMPACT can re-plan to offer another suitable trajectory from the current position. (b) Level

3 Autonomy: The USV is controlled by CoMPACT along a planned trajectory to complete the inspection mission without colliding into

known obstacles. Pop-up threats are still possible hindrances, and the operator is responsible for re-planning given such interruptions.

Since CoMPACT instructs the vessels to travel in tandem when under autonomous control (Level 3, 4, and 5), theresults for these cases look the same for a single vessel or multiple vessels. The manual control cases, on the other

hand, allow the operator to control the USVs individually or in tandem. Figure 25(a), illustrates how the operator maychose to control the USVs individually during a portion of the task. After each USV has cleared the island, they are

again synchronized so the blue USV assumes a position in formation with the green vessel.

For Level 2 autonomy, CoMPACT generates only one trajectory, as it is assumed valuable for both USVs to follow.Furthermore, with the capability of having the operator assume direct control of only one of the USVs and relying

upon autonomous formation control for the other USV lightens the burden of the operator through the course of the

trajectory.

Figures 25(b) and 26 show details of two USVs tasked to inspect the suspicious vessel, and the representative termi-

nal trajectory they are tasked with. The vessels are initially being controlled autonomously in a formation configuration,and then they break into individual courses to inspect the vessel. In this way, the USVs cooperate autonomously to

accomplish the ship inspection task.

IX. Conclusions

This paper presents research and development of an effective Collaborative Mission Planning, Autonomy and Con-trol Technology (CoMPACT) for multiple Unmanned Surface Vehicles. CoMPACT enables autonomous collaboration,

21 of 23


(a) (b)

Figure 25. (a) Level 0 Autonomy– Two USVs: In this simulation, the USVs are controlled individually around an island, being navigated

through significantly different paths. (b) Level 3 Autonomy– Two USV Inspection: In this simulation, the USVs are controlled autonomously

through the obstacle field, then tasked to inspect the suspicious vessel.

Figure 26. Level 3 Autonomy– Two USV Inspection: The two USVs work in concert to perform the inspection. One USV crosses ahead of

the suspected target to survey the port side, while the other remains on the starboard side.

coordination & control, and is characterized by a versatile system architecture for different methods of collaborative

decision making (e.g. centralized, semi- and fully-decentralized). CoMPACT achieves effective conflict resolution,

dynamic mission/mode/task re-planning, and efficient vehicle-to-target assignment using Finite State Machine (FSM)based task execution (state or event-based transitions).

A major benefit of CoMPACT is its ability to decrease operator workload while increasing mission efficiency.

Furthermore, the technology is modular in design which allows straightforward expansion, and is well suited formultiple heterogeneous USVs or other types of unmanned vehicles (UAV, UUV, UGV) within the overall architecture.

CoMPACT integrates a conceptual solution of a user-friendly operator interface plug-in that enables quick tran-sitions between varying levels of autonomy. The approach is strengthened by its highly efficient path planning us-

ing modified Rapidly-exploring Random Trees (RRT) algorithm, effective theater of operation representation using

Quadtrees and grid techniques, vehicle assignment algorithms based on a suitably chosen capability matrix and RRT,highly efficient pursuit algorithms for moving targets through obstacle fields. Other strengths of the research presented

here are that higher-fidelity models, communication and sensor models can readily be incorporated into the overall

architecture.

The CoMPACT system has been successfully implemented within several complex maritime mission scenarios

involving multiple coordinated USV platoons and including area search, mine countermeasures, and anti-submarine

warfare.

Acknowledgments

The authors would like to thank Louis Larkin and Eric Holmes from Lockheed Martin Undersea and Security Sys-

tems for their technical guidance and support for this project, including scenario definitions and concept of operations.

22 of 23


References1J. D. Boskovic, J. Redding and R. K. Mehra, “Decentralized Collaborative Planning & Control for Target Interception using a Network of

Autonomous Underwater Vehicles (AUVs)”, Final Report for DARPA Phase I SBIR, Contract No. W31P4Q-06-C-0024, April 2006.2R. Beard, T. McLain, M. Goodrich and E. Anderson, “Coordinated target assignment and intercept for unmanned air vehicles”, IEEE Trans.

Robotics and Automation, 18(5), 911- 922, Dec. 2002.3A. Ryan, M. Zennaro, A. Howell, R. Sengupta and J. K. Hedrick, “An Overview of Emerging Results in Cooperative UAV Control”, in Proc.

43rd IEEE Conference on Decision and Control December 14-17, 2004 Atlantis, Paradise Island, Bahamas.4A. Richards, J. Bellingham, M. Tillerson, and J. How, “Coordination and Control of Multiple UAVs,” Proc. AIAA Guidance, Navigation and

Control Conference, Monterey, CA, Aug 2002.5D. Scheidt, J. Stipes and T. Neighoff, “Cooperating Unmanned Vehicles”, Proc. 2005 IEEE International Conf. on Networking, Sensing &

Control, pp. 326-331, Tuscon, AZ, March 19-22, 20056R. W. Chalmers, D. H. Scheidt, T. M. Neighoff, S. J. Witwicki and R. J. Bamberger, “Cooperating Unmanned Vehicles”, AIAA 2004-6252,

in Proc. AIAA 1st Intelligent Systems Technical Conference, 20 - 22 September 2004, Chicago, Illinois.7R. E. Burkard, “Selected topics in assignment problems,” Technical Report, Institute of Mathematics B, Technical University of Graz,

November 1999.8P. Stone and M. Veloso, “Task decomposition, dynamic role assignment, and low-bandwidth communication for real-time strategic team-

work,” Artificial Intelligence, 110(2), 241-273, June 1999.9F. Brandt, W. Brauer, and G. Weiss, “Task assignment in multiagent systems based on Vickrey-type auctioning and leveled commitment

contracting,” in Proc. Cooperative Information Agents (CIA-2000) Workshop, Boston, MA, 2000, pp. 95-106.10Y. Li, P. M. Pardalos, and M. G. C. Resende, “A greedy randomized adaptive search procedure for the quadratic assignment problem,”

DIMACS Series on Discrete Math. and Theoretical Comp. Sci., vol. 16, pp. 237-261, 1994.11“The Navy Unmanned Surface Vehicle (USV) Master Plan”, www.navy.mil/navydata/technology/usvmppr.pdf, 23 July 2007.12“The Navy Unmanned Undersea Vehicle (USV) Master Plan”, www.navy.mil/navydata/technology/uuvmp.pdf, 9 November 2004.13S. M. LaValle, “From dynamic programming to RRTs: Algorithmic design of feasible trajectories,” In A. Bicchi, H. I. Christensen, and D.

Prattichizzo, editors, Control Problems in Robotics, pages 19–37. Springer-Verlag, Berlin, 2002.14S. M. LaValle, “Rapidly-exploring random trees: A new tool for path planning,” TR 98-11, Computer Science Dept., Iowa State University,

Oct. 1998.15J. D. Boskovic, J. Amin, J. Redding, R. K. Mehra, Y. Kang and K. Hedrick, “Multi-layer Architecture for Trajectory Replanning and

Intelligent plan eXecution (MATRIX)”, Final Report for NASA Ames Phase II STTR, Contract No. NNA04AA63C, May 2006.16J. Amin, J. Boskovic and R. Mehra, “A Fast and Efficient Approach to Path Planning for Unmanned Vehicles”, Proc. AIAA Guidance,

Navigation & Control Conference, Keystone, CO, Aug. 2006.17J. D. Boskovic, J. Redding and K. Chan, “Heterogeneous Collaborative Control of Unmanned Aerial Vehicles”, Final Report for Northrop

Grumman Subcontract No. 2632876, Dec. 2006.18J. D. Boskovic, J. Redding, R. K. Mehra, S. Sastry, H. Chung and D. Shim, “Pilot-Directed Computer Assisted Helicopter Formation Flying”,

Final Report for Army Research Office Phase I SBIR, Contract No. W911NF-05-C-0088, February 2006.19J. D. Boskovic, R. Prasanth, J. Byrne, D. Garagic, J. Amin, S. Bergstrom, M. Cosgrove, and R. K. Mehra, “Development of Intelligent

Model Predictive Control Algorithms and a Software Design Toolbox for Autonomous Systems”, Final Report for DARPA Phase II SBIR, ContractNo. DAAH01-00-C-R187, March 2005.

20J. D. Boskovic, R. Prasanth and R. K. Mehra, “A Multi-Layer Autonomous Intelligent Control Architecture for Unmanned Aerial Vehicles”,AIAA Journal of Aerospace Computing, Information, and Communication (JACIC), Special Issue on Intelligent Systems, Vol. 1, pp. 605-628, Dec.2004.

21J. D. Boskovic, S.-M. Li, R. Prasanth, M. Donahue and R. K. Mehra, “Higher Level Supervision, Inference and Decision Algorithms andSoftware Tool Development for Multiple Autonomous UCAVs”, Final Report for ONR SBIR Phase II, March 2002.

22P. van Oosterom and T. Vijlbrief, “The Spatial Location Code”, in Proceedings of the Seventh International Symposium on Spatial DataHandling, Delf University, August 1996.

23H. Samet, The Design and Analysis of Spatial Data Structures, Addison-Wesley, Reading, MA, 1990.24E. W. Dijkstra, “A Note on Two Problems in Connection with Graphs”, Numerische Mathematik, Vol. 1, pp. 269-271, 1959.25T. H. Cormen, C. E. Leiserson, R. L. Rivest, and C. Stein, Introduction to Algorithms, Second Edition, MIT Press and McGraw-Hill, 2001.

Section 24.3: Dijkstra’s algorithm, pp.595-601.26J. D. Boskovic, R. Prasanth and R. K. Mehra, “A Multi-Layer Autonomous Intelligent Control Architecture for Unmanned Aerial Vehicles”,

AIAA Journal of Aerospace Computing, Information, and Communication (JACIC), Special Issue on Intelligent Systems, Vol. 1, pp. 605-628,December 2004.

27J. D. Boskovic, Ravi Prasanth, J. Byrne, D. Garagic, Jayesh Amin, S. Bergstrom, M. Cosgrove, and R. K. Mehra, “Development of IntelligentModel Predictive Control Algorithms and a Software Design Toolbox for Autonomous Systems”, Final Report for DARPA Phase II SBIR, ContractNo. DAAH01-00-C-R187, March 2005.

28C. Ericson, Real-Time Collision Detection, Morgan Kaufmann Publishers, San Francisco, CA, 2005 (www.mkp.com).29J. Howlett, G. Schulein and M. H. Mansour, “A Practical Approach to Obstacle Field Route Planning for Unmanned Rotorcraft”, Presented

at the American Helicopter Society 60th Annual Forum, Baltimore, MD, June 7-10, 2004.30J. R. Miller, “A 3D Color Terrain Modeling System for Small Autonomous Helicopters,” Technical Report No. CMU-RI-TR-02-07, Robotics

Institute, Carnegie Mellon University, February 2002.31J. R. Miller and O. Amidi, “3-D Site Mapping with the CMU Autonomous Helicopter,” in Proceedings of the 5th International Conference

on Intelligent Autonomous Systems, June 1998.32J. Guivant and E. Nebot, “Optimization of the Simultaneous Localization and Map Building Algorithm for Real Time Implementation” IEEE

Transactions of Robotics and Automation, May 2001.33S. Thrun, M. Diel and D. Hahnel, “Scan Alignment and 3-D Surface Modeling with a Helicopter Platform”, The 4th International Conference

on Field and Service Robotics, July 14-16, 2003.34S.J. Russell and P. Norvig, Artificial Intelligence: A Modern Approach (2nd ed.), Upper Saddle River, NJ: Prentice Hall, ISBN 0-13-790395-

2, aima.cs.berkeley.edu, 2003.35D. Poole, David; A. Mackworth and R. Goebel, Computational Intelligence: A Logical Approach, Oxford University Press, 1998.

23 of 23


Documents

Collaborative Mission Planning, Autonomy and Control ... · Collaborative Mission Planning, Autonomy and Control Technology (CoMPACT) for Unmanned Surface Vehicles ∗ Joshua Redding†,