Structured Control for Autonomous Robots

Reid G. SimmonsCarnegie Mellon University

Uday Rajanna

Task Control Architecture:

1) Layering reactive behaviours onto deliberative components

2) It is a structured control approach where the deliberative components handle normal situations and the reactive behaviours handle exceptional situations.

3) TCA control constructs facilitates modular and evolutionary system development.

Task Control refers to the problem of coordinating perceptual, planning and execution components of a robot system to achieve a given set of goals

Behaviour Based Systems:

As the number of behaviours and the degree of interaction between them grows, it becomes increasingly difficult to design good arbitration schemes of which behaviour takes a higher or lower priority of execution.

TCA Control Constructs:

1) Distributed inter-process communication2) Task decomposition and temporal constraints between subtasks.3) Resource allocation and management.4) Execution monitoring5) Exception handling.

TCA class of messages:

1) Inform2) Query3) Goal4) Command5) Monitor6) Exception

Ambler walking system:

1) Central Control2) Gait Planner3) Footfall planner4) Leg recovery planner5) Error recovery planner6) Scanner Interface7) Image Queue manager8) Local terrain mapper9) Message routing table10) Resource Schedules11) Task Trees12) Real time controller

TCA provides mechanisms for

* Deliberation Uses a task tree that keeps the overall process in mind. Constraints

like delay planning, handling intervals, achievement interval, goal interval are used.

* Reactivity Uses task trees which are reactive in nature in conjunction with the deliberation scheme to account for sudden changes in environment and exception errors.

TCA also allows for incremental development of the system with minimal impact to the existing architecture.

TCA module utilization was improved to be very concurrent by the addition of DP between “Take Steps” node and the node directly preceding its parent.

Conclusions:

1) TCA provides a framework for developing and controlling autonomous robots

2) The control of planning, perception and action must be well structured for general purpose robots to succeed in rich and uncertain environments.

3) TCA supports both deliberative and reactive architectures with the side coming into play in case of exceptions and changes in environments.

4) Use of constraints provides a basis for analysing interactions between behaviours.

5) Using formal methods to analyse constraints on behaviour would greatly facilitate development of robot systems.

6) Incremental addition of new modules can be performed.