sorinng

Embed Size (px)

Citation preview

  • 7/29/2019 sorinng

    1/41

  • 7/29/2019 sorinng

    2/41

    Physicomimetics forSwarm Formations andObstacle Avoidance

    Suranga Hettiarachchi Ph.D.

    Computer Science and Multimedia

    Eastern Oregon University

    Funded by Joint Ground Robotics Enterprise - DOD

  • 7/29/2019 sorinng

    3/41

    Focus of the Talk

    Improved performance in swarm obstacleavoidance: Scales to far higher numbers of robots and

    obstacles than the norm Hardware Implementation

    Implemented obstacle avoidance algorithm onreal robots

    ObstacleAvoidance

    HardwareImplementation

    Simulated RobotSwarms

  • 7/29/2019 sorinng

    4/41

    Outline

    Robot Swarms

    Physicomimetics Framework

    Swarm Learning

    Obstacle Avoidance with Physical Robots

    Conclusion and Future Work

  • 7/29/2019 sorinng

    5/41

    Robot Swarms

    Robot swarms can act as distributedcomputers, solving problems that asingle robot cannot

    For many tasks, having a swarmmaintain cohesiveness while avoidingobstacles and performing the task is ofvital importance

    Example :Chemical Plume

    Source Tracingpicture: Maxelbots at UW-DRL

  • 7/29/2019 sorinng

    6/41

    Swarm Advantages

    Swarms of robots are effective:

    They can perform tasks that one expensiverobot cannot.

    Example: UAVs for surveillance Swarms are robust:

    Even if some robots fail, the swarm can stillachieve the task.

    Robots can be reused:

    Functionally specific agents can be used tosolve different problems

    picture: global aircraft

  • 7/29/2019 sorinng

    7/41

    Outline

    Robot Swarms

    Physicomimetics Framework

    Swarm Learning

    Obstacle Avoidance with Physical Robots

    Conclusion and Future Work

  • 7/29/2019 sorinng

    8/41

    Physicomimetics for Robot Control

    Biomimetics: Gain inspiration frombiological systems and ethology.

    Physicomimetics: Gain inspiration from

    physical systems. Good for formations.

  • 7/29/2019 sorinng

    9/41

    Why we mimic physics?

    Aggregate behaviors seen in classical physics ispotentially reproducible with collections ofmobile agent.

    Incorporate our understanding of classicalphysics to derive collective behavior of robots.

    We are not restricted to copying physicsprecisely, so modifications are possible.

  • 7/29/2019 sorinng

    10/41

    PhysicomimeticsFramework

    Robots have limited sensor range,

    and friction for stabilization

    FF

    FFF

    Virtual forces F on a robot A by other

    robots ai and the environment cause a

    d displacement in its behavior.

    d

    a1a

    2

    a3

    a4

    A

    Environment

    Robots are controlled

    via virtual forces

    from nearby robots,

    goals, and obstacles.

    F = ma control law.

    Seven robots form a hexagon

  • 7/29/2019 sorinng

    11/41

    Two Classes of Force Laws

    p

    ji

    r

    mGmF

    7

    6

    13

    12224

    r

    c

    r

    dF

    The left Newtonian force law, is good for creatingswarms in rigid formations. The right Lennard-Jones force law (LJ) more easily models fluid

    behavior, which is potentially better for maintaining

    cohesion while avoiding obstacles.

    The classic law Novel use of LJ force law for robot control

  • 7/29/2019 sorinng

    12/41

    What do these forcelaws look like?

    Change in Force MagnitudeWith Varying Distance forRobot Robot Interactions

    Fmax = 1.0

    Fmax = 4.0

    Desired Robot Separation Distance = 50

  • 7/29/2019 sorinng

    13/41

    Outline

    Robot Swarms

    Physicomimetics Framework

    Swarm Learning

    Obstacle Avoidance with Physical Robots

    Conclusion and Future Work

  • 7/29/2019 sorinng

    14/41

    Swarm Learning

    Typically, the interactions between the swarmmembers are learned via simulation.

    Swarm Simulation

    Initial RulesFinal Rules

    that achieve thedesired behavior

    Evolutionary Algorithm(EA)

    FitnessRules

  • 7/29/2019 sorinng

    15/41

    Swarm Simulation Environment

  • 7/29/2019 sorinng

    16/41

    Learning Approach

    An Evolutionary Algorithm (EA) is used to evolvethe rules for the robots in the swarm.

    A global observer assigns fitness to the rules

    based on the collective behavior of the swarm inthe simulation.

    Each member of the swarm uses the same rules.The swarm is a homogeneous distributed

    system. For physicomimetics, the rules are vectors

    of force law parameters.

  • 7/29/2019 sorinng

    17/41

    Force Law Parameters

    Parameters of the Newtonian force lawG-gravitational constant of robot-robot interactionsP- power of the force law for robot-robot interactionsFmax- maximum force of robot-robot interactions

    Similar 3-tuples for obstacle/goal-robot interactions.

    Parameters of the LJ force law- strength of the robot-robot interactionsc- non-negative attractive robot-robot parameter

    d- non-negative repulsive robot-robot parameterFmax- maximum force of robot-robot interactions

    Similar 4-tuples for obstacle/goal-robot interactions.

    Gr-r Pr-r Fmaxr-r Gr-o Pr-o Fmaxr-o Gr-g Pr-g Fmaxr-g

    r-r cr-r dr-r Fmaxr-r r-o cr-o dr-o Fmaxr-o

    r-g cr-g dr-g Fmaxr-g

  • 7/29/2019 sorinng

    18/41

    Measuring Fitness

    Connectivity (Cohesion) : maximum numberof robots connected via a communication path.

    Reachability (Survivability) : percentage of

    robots that reach the goal. Time to Goal : time taken by at least 80% of

    the robots to reach the goal.

    goalconnectivity4R

    reachability

    High fitness corresponds to high connectivity,high reachability, and low time to goal.

  • 7/29/2019 sorinng

    19/41

    Connectivity of Robots

  • 7/29/2019 sorinng

    20/41

  • 7/29/2019 sorinng

    21/41

    ForceLaw

    Robots Obstacles

    20 40 60 80 100

    Newt20 1160 1260 1290 1530 1920

    100 - - - - -

    LJ

    20 470 480 490 510 520

    100 640 650 670 680 690

    Time for 80% of the Robotsto Reach the Goal

  • 7/29/2019 sorinng

    22/41

    Summary of Results

    We compared the performance of the bestNewtonian force law found by the EA to the best LJforce law.

    The Newtonian force law produces more rigidstructures making it difficult to navigate throughobstacles. This causes poor performance, despitehigh connectivity.

    LJ is superior, because the swarm acts as aviscous fluid. Connectivity is maintained while

    allowing the robots to reach the goal in a timelymanner.

    The LJ force law demonstrates scalability in thenumber of robots and obstacles.

  • 7/29/2019 sorinng

    23/41

    Outline

    Robot Swarms

    Physicomimetics Framework

    Swarm Learning

    Obstacle Avoidance with Physical Robots

    Conclusion and Future Work

  • 7/29/2019 sorinng

    24/41

    Obstacle Avoidancewith Robots

    Use three Maxelbot robots

    Use 2D trilateration localization

    algorithm (Not a part of this talk)

    Design and develop obstacle avoidance module(OAM)

    Implement physicomimetics on a real outdoorrobot

  • 7/29/2019 sorinng

    25/41

    Hardware Architectureof Maxelbot

    MiniDRAGON formotor control,

    executesPhysicomimetics

    MiniDRAGON fortrilateration,

    provides robotcoordinates

    OAMAtoD conversion

    RF and acoustic sensors

    IR sensors

    I2C

    I2C

    I2C

  • 7/29/2019 sorinng

    26/41

    Formation ControlMethodology

    Measure the quality of AP-lite withoutrepulsions from obstacles

    All experiments are conducted outdoor

    Three Maxelbots: One leader and two followers Results averaged over five runs

    Leader remotely controlled (NO AP-lite)

    Robots DO NOT have obstacle avoidance

    capability Focus is on the formation control, not the

    obstacle avoidance

  • 7/29/2019 sorinng

    27/41

    Why AP-lite?

    Capable of maintaining formations of robots

    Designed as a leader-follower algorithm

    Allows robots to move quickly, due to minimalcommunication

    Can use theory to set parameters

  • 7/29/2019 sorinng

    28/41

    Triangular Formation

  • 7/29/2019 sorinng

    29/41

  • 7/29/2019 sorinng

    30/41

    Linear Formation

  • 7/29/2019 sorinng

    31/41

  • 7/29/2019 sorinng

    32/41

    Physicomimetics forObstacle Avoidance

    Constant virtual attractive goal force in frontof the leader

    Virtual repulsive forces from four sensors

    mounted on the front of the leader, if obstaclesdetected

    The resultant force creates a change in velocitydue to F = ma

    Power supply to motors are changed based onthe forces acting on the leader.

  • 7/29/2019 sorinng

    33/41

    Obstacle AvoidanceMethodology

    Measure the performance of physicomimeticswith repulsion from obstacles

    All experiments are conducted outdoor

    Three Maxelbots: One leader and two followers

    Graphs show the correlation between rawsensor readings and motor power

    Leader uses the physicomimetics algorithmwith the obstacle avoidance module

    Focus is on the obstacle avoidance by theleader, not the formation control

  • 7/29/2019 sorinng

    34/41

    Maxelbot Turning Left - Obstacle on the Right

    -100

    0

    100

    200

    300

    400

    500

    600

    700

    800

    1 1001 2001 3001 4001 5001 6001 7001 8001 9001

    Time

    SensorReadinga

    ndMotorPower

    Right-most Sensor Reading

    Power to Left Motor

    If there is an obstacle on the right, power to left motor is reduced

  • 7/29/2019 sorinng

    35/41

    If there is an obstacle in front, power to both motors is reduced

    Maxelbot Stopping Behavior - Both Middle Sensors Detect an Obstacle

    0

    100

    200

    300

    400

    500

    600

    1 1001 2001 3001 4001 5001 6001 7001 8001 9001

    Time

    S

    ensorReadinga

    andMotorPow

    er

    Ave. of the Two Middle Sensors

    Ave. of the Motor Power

  • 7/29/2019 sorinng

    36/41

    Further Analysis of SensorReading and Motor Power

    Scatter plots show how much one variable is affectedby the other

    Provide a broader picture of change in motor powerwhen the robot sensors detects obstacles

    Shows the correlation of motor power with distance

    to an obstacle in inches (the robots ignore obstaclesgreater than 30 away)

  • 7/29/2019 sorinng

    37/41

    Maxelbot Turning Left - Obstacle on the Right

    -20

    -10

    0

    10

    20

    30

    40

    50

    60

    70

    80

    0 10 20 30 40 50 60 70 80 90 100

    Distance to Obstacle (inches)

    PowertoLeftMotor

    Lag in starting due to AP inertia.Helps counteract noisy sensors.

    Lag in stopping due to AP inertia.Helps counteract noisy sensors.

    Right sensorsees obstacle

    Right middle sensorsees obstacle

  • 7/29/2019 sorinng

    38/41

    Maxelbot Stopping Behavior - Both Middle Sensors Detect an

    Obstacle

    0

    10

    20

    30

    40

    50

    60

    70

    80

    0 10 20 30 40 50 60 70 80 90 100

    Distance to Obstacle (inches)

    AverageofLeftand

    RightMotorPower

    Power will be reduced if theoutermost sensors see anobstacle when the inner

    sensors do not.

  • 7/29/2019 sorinng

    39/41

    Outline

    Robot Swarms

    Physicomimetics Framework

    Swarm Learning

    Obstacle Avoidance with Physical Robots

    Conclusion and Future Work

  • 7/29/2019 sorinng

    40/41

    Future Work

    Provide obstacle avoidance capability to all therobots in the formation Develop robots with greater data exchangecapability

    Adapt the physicomimetics framework toincorporate performance feedback for specifictasks and situational awareness Extend the physicomimetics framework forsensing and performing tasks in a marine

    environment (with Harbor Branch) Introduce robot/human roles and interactions todistributed evolution architecture

  • 7/29/2019 sorinng

    41/41

    Thank You

    Questions?

    Movie of 3 Maxelbots,Leader has OAM