Behavior-based Behavior-based Multirobot Multirobot Architectures Architectures
Behavior-based Behavior-based Multirobot Multirobot
ArchitecturesArchitectures
Why Behavior Based Control Why Behavior Based Control for Multi-Robot Teams?for Multi-Robot Teams?
Multi-Robot control naturally grew Multi-Robot control naturally grew out of single robot controlout of single robot control– Reactive: No state informationReactive: No state information– Planner: State space is already hugePlanner: State space is already huge
Adding Adding nn additional robots to a state space additional robots to a state space of of s s results in an state space of results in an state space of ssnn
– Hybrid: Same problems as a planning Hybrid: Same problems as a planning control systemcontrol system
Why Behavior Based Control Why Behavior Based Control for Multi-Robot Teams?for Multi-Robot Teams?
Behavior Based ControlBehavior Based Control Pros:Pros:
– Since control is locally situated it scales Since control is locally situated it scales wellwell
– No reliance on global communication or No reliance on global communication or planning results in robots better able to planning results in robots better able to handle sensor and actuator noise handle sensor and actuator noise
– Primitive Behaviors are relatively simplePrimitive Behaviors are relatively simple
Why Behavior Based Control Why Behavior Based Control for Multi-Robot Teams?for Multi-Robot Teams?
Cons:Cons:– Difficult to createDifficult to create
ExperimentalExperimental
– Difficult to analyzeDifficult to analyze Actions of robots depend on the actions of Actions of robots depend on the actions of
other robotsother robots Behavior of the team is based on the Behavior of the team is based on the
interactions between robots instead of an interactions between robots instead of an individual robots control strategyindividual robots control strategy
Issues in Behavior Based Multi-Issues in Behavior Based Multi-Robot ControlRobot Control
How to create and combine How to create and combine behaviors to accomplish a given behaviors to accomplish a given goal?goal?
How to coordinate robot behaviors?How to coordinate robot behaviors?– Use Communication?Use Communication?– What kind of knowledge should the What kind of knowledge should the
team have?team have? Purely local control?Purely local control? Hybrid local and global control?Hybrid local and global control?
Behavior Creation and Behavior Creation and SelectionSelection
Bottom-UpBottom-Up– Primitive behaviors should be minimalist in Primitive behaviors should be minimalist in
that sense that a primitive behaviors can that sense that a primitive behaviors can not be derived from other primitive not be derived from other primitive behaviorsbehaviors
– Constrained by the robot’s physical Constrained by the robot’s physical capabilitiescapabilities
– Constrained by the environmentConstrained by the environment Top-DownTop-Down
– Behaviors are constrained by the types of Behaviors are constrained by the types of goals the must be accomplished by a teamgoals the must be accomplished by a team
Test CasesTest Cases
EquipmentEquipment– 20 mobile robots equipped with infra-red 20 mobile robots equipped with infra-red
sensors, micro-switches, sonar, and sensors, micro-switches, sonar, and radioradio
EvaluationEvaluation– RepeatabilityRepeatability– StabilityStability– RobustnessRobustness– ScalabilityScalability
Test Cases ContinuedTest Cases Continued
Primitive BehaviorsPrimitive Behaviors– AvoidanceAvoidance– FollowingFollowing– AggregationAggregation– DispersionDispersion– HomingHoming– WanderingWandering– Grasping / DroppingGrasping / Dropping
Test Cases ContinuedTest Cases Continued
FlockingFlocking– Summation ofSummation of
AvoidanceAvoidance AggregationAggregation WanderingWandering
– Addition of Homing for goal directed behaviorAddition of Homing for goal directed behavior ResultsResults
– Goal directed behavior without dependence on Goal directed behavior without dependence on a leader and robust in case of single robot a leader and robust in case of single robot failurefailure
– FlockingFlocking
Test Cases ContinuedTest Cases Continued
ForagingForaging– Temporally switch between: avoidance, Temporally switch between: avoidance,
dispersion, following, homing, and dispersion, following, homing, and wanderingwandering
ResultsResults– Basic behaviors were empirically shown Basic behaviors were empirically shown
to be robust and flexible in collecting to be robust and flexible in collecting pucks and dropping them off at a goal pucks and dropping them off at a goal locationlocation
ReferenceReference
Mataric, Mataric, Issues and Approaches in Issues and Approaches in the Design of Collective Autonomous the Design of Collective Autonomous AgentsAgents
Behavior Based Multi-Robot Behavior Based Multi-Robot Team CoordinationTeam Coordination
CommunicationCommunication– Often times relies on Master-Slave HierarchyOften times relies on Master-Slave Hierarchy
Inherent brittleness to this approachInherent brittleness to this approach– Bandwidth limitationsBandwidth limitations– Robustness - Master failure?Robustness - Master failure?– Heterogeneous or Homogeneous approach?Heterogeneous or Homogeneous approach?
– Is explicit communication needed or is Is explicit communication needed or is implicit communication enough to achieve implicit communication enough to achieve the goal?the goal?
– If communications are used how much is If communications are used how much is needed and what should be communicated?needed and what should be communicated?
Cooperation Without Cooperation Without CommunicationCommunication
Is cooperation in a behavior based Is cooperation in a behavior based multi-robot team without multi-robot team without communication possible?communication possible?
If so, how effective is it?If so, how effective is it?
Behavioral CompositionBehavioral Composition ForageForage
– NoiseNoise– Avoid static obstaclesAvoid static obstacles– Avoid robotsAvoid robots
AcquireAcquire– Move to goalMove to goal– Avoid static obstaclesAvoid static obstacles– NoiseNoise
DeliverDeliver– Move to goalMove to goal– Avoid static obstaclesAvoid static obstacles– Avoid robotsAvoid robots– NoiseNoise
Test CasesTest Cases SimulationSimulation Map Size: 64 x 64 unitsMap Size: 64 x 64 units Maximum Sensor Distance: 25 unitsMaximum Sensor Distance: 25 units ForageForage
– Noise Gain: 1.2Noise Gain: 1.2– Noise Persistence: 4Noise Persistence: 4– Avoid Obstacles Gain: 1.0Avoid Obstacles Gain: 1.0– Avoid Robots Gain: 0.5Avoid Robots Gain: 0.5
Acquire/DeliverAcquire/Deliver– Noise Gain: 0.2Noise Gain: 0.2– Noise Persistence: 2Noise Persistence: 2– Move to Goal Gain: 1.0Move to Goal Gain: 1.0– Avoid Robots Gain: 0.1Avoid Robots Gain: 0.1
ResultsResults
2 Robots / 1 2 Robots / 1 AttractorAttractor
ResultsResults
4 robots / 4 4 robots / 4 attractorsattractors
ResultsResults
Without using communication, the Without using communication, the simulation still shows coherent simulation still shows coherent cooperation between the team cooperation between the team membersmembers– Cheaper hardwareCheaper hardware– Fewer points of failureFewer points of failure
ReferencesReferences
Arkin, Arkin, Cooperation without Cooperation without communicationcommunication
For more quantitative comparisons For more quantitative comparisons between levels of communication between levels of communication see: Balch/Arkin,see: Balch/Arkin, Communication in Communication in Reactive Multiagent Robotic SystemsReactive Multiagent Robotic Systems
Behavior Based Multi-Robot Behavior Based Multi-Robot Team Coordination ContinuedTeam Coordination Continued
Local versus Global Control LawsLocal versus Global Control Laws– Local ControlLocal Control
Simple and contain emergent propertiesSimple and contain emergent properties Oftentimes unclear as to how to design local Oftentimes unclear as to how to design local
control lawscontrol laws
– Global ControlGlobal Control Allow for more coherent team cooperationAllow for more coherent team cooperation Often results in increased communication Often results in increased communication
requirementsrequirements
Global Control LawsGlobal Control Laws Global Goal KnowledgeGlobal Goal Knowledge
– Information concerning the overall goal of the agents Information concerning the overall goal of the agents behaviorbehavior
– Can be encoded into robot if the goal is not dynamicCan be encoded into robot if the goal is not dynamic Global KnowledgeGlobal Knowledge
– Information concerning what other robots are doingInformation concerning what other robots are doing– Information concerning what other robots will doInformation concerning what other robots will do
Obtaining this knowledge must often come from Obtaining this knowledge must often come from outside sourcesoutside sources
The knowledge is computationally costlyThe knowledge is computationally costly Oftentimes all the needed global knowledge is not Oftentimes all the needed global knowledge is not
knownknown
Local Control LawsLocal Control Laws
Computationally SimpleComputationally Simple Handle dynamic environments wellHandle dynamic environments well Oftentimes do not produce optimal Oftentimes do not produce optimal
resultsresults Must rely on physical sensorsMust rely on physical sensors
ExperimentExperiment
Simulation of mission Simulation of mission involving formation involving formation maintenance while maintenance while moving to goalmoving to goal
4 different strategies 4 different strategies with increasing global with increasing global controlcontrol
Quantitatively Quantitatively measured via measured via deviation from the deviation from the formation and time formation and time taken to reach goaltaken to reach goal
Experiment ContinuedExperiment Continued
Strategy I: Local Strategy I: Local Control OnlyControl Only– Effective for smooth Effective for smooth
trajectoriestrajectories– Sharp turns cause Sharp turns cause
formation to break formation to break up due to local up due to local controlcontrol Robots maintain their Robots maintain their
position by staying a position by staying a fixed distance from a fixed distance from a certain side of certain side of neighborneighbor
Experiment ContinuedExperiment Continued
Strategy II: Local Strategy II: Local Control Augmented by Control Augmented by Global GoalGlobal Goal– Robots given Robots given
knowledge of global knowledge of global goal: Maintain line goal: Maintain line formationformation
– Robot D now moves to Robot D now moves to a more globally a more globally appropriate positionappropriate position
– May be inappropriate if May be inappropriate if B is merely avoiding B is merely avoiding obstacleobstacle
Experiment ContinuedExperiment Continued
Strategy III: Local Strategy III: Local Control Augmented Control Augmented with Global Goal with Global Goal and Partial Global and Partial Global InformationInformation– At time of robot B’s At time of robot B’s
turn, other robots turn, other robots are informed of are informed of destination of destination of waypoint Xwaypoint X
Experiment ContinuedExperiment Continued
Strategy IV: Local Strategy IV: Local control augmented by control augmented by global goal and more global goal and more complete global complete global informationinformation– Robots are given Robots are given
complete knowledge of complete knowledge of leaders routeleaders route
– Allows other robots to Allows other robots to predict future positions predict future positions of the leader and of the leader and resulting positions for resulting positions for themselvesthemselves
ResultsResults As global information As global information
increases formation increases formation error and completion error and completion time decreases time decreases
Global goals are useful Global goals are useful to incorporate if goals to incorporate if goals are known at run timeare known at run time
Global information is Global information is useful in static, well useful in static, well defined environmentsdefined environments
Results ContinuedResults Continued Local control in situations where Local control in situations where
accomplishing the task as opposed to how accomplishing the task as opposed to how the task accomplished often provides a the task accomplished often provides a suitable approximation of optimal behaviorsuitable approximation of optimal behavior
Behavioral analysis in local control my Behavioral analysis in local control my approximate global knowledgeapproximate global knowledge
General Rule: “General Rule: “Local control information Local control information should be used to ground global knowledge should be used to ground global knowledge in the current situation. This allows the in the current situation. This allows the agents to remain focused on the overall agents to remain focused on the overall goals of their group while reacting to the goals of their group while reacting to the dynamics of their current situation.dynamics of their current situation.””
ReferencesReferences
Parker, Parker, Designing Control Laws for Designing Control Laws for Cooperative Agent TeamsCooperative Agent Teams
QuestionsQuestions
??????