Behaviorbased Multirobot Architectures Why Behavior Based Control for
Behavior-based Multirobot Architectures
Why Behavior Based Control for Multi-Robot Teams? n Multi-Robot control naturally grew out of single robot control – Reactive: No state information – Planner: State space is already huge § Adding n additional robots to a state space of s results in an state space of sn – Hybrid: Same problems as a planning control system
Why Behavior Based Control for Multi-Robot Teams? n Behavior Based Control n Pros: – Since control is locally situated it scales well – No reliance on global communication or planning results in robots better able to handle sensor and actuator noise – Primitive Behaviors are relatively simple
Why Behavior Based Control for Multi-Robot Teams? n Cons: – Difficult to create § Experimental – Difficult to analyze § Actions of robots depend on the actions of other robots § Behavior of the team is based on the interactions between robots instead of an individual robots control strategy
Issues in Behavior Based Multi. Robot Control n How to create and combine behaviors to accomplish a given goal? n How to coordinate robot behaviors? – Use Communication? – What kind of knowledge should the team have? § Purely local control? § Hybrid local and global control?
Behavior Creation and Selection n Bottom-Up – Primitive behaviors should be minimalist in that sense that a primitive behaviors can not be derived from other primitive behaviors – Constrained by the robot’s physical capabilities – Constrained by the environment n Top-Down – Behaviors are constrained by the types of goals the must be accomplished by a team
Test Cases n Equipment – 20 mobile robots equipped with infra-red sensors, micro-switches, sonar, and radio n Evaluation – Repeatability – Stability – Robustness – Scalability
Test Cases Continued n Primitive Behaviors – Avoidance – Following – Aggregation – Dispersion – Homing – Wandering – Grasping / Dropping
Test Cases Continued n Flocking – Summation of § § § Avoidance Aggregation Wandering – Addition of Homing for goal directed behavior n Results – Goal directed behavior without dependence on a leader and robust in case of single robot failure – Flocking
Test Cases Continued n Foraging – Temporally switch between: avoidance, dispersion, following, homing, and wandering n Results – Basic behaviors were empirically shown to be robust and flexible in collecting pucks and dropping them off at a goal location
Reference Issues and Approaches in the Design of Collective Autonomous Agents n Mataric,
Behavior Based Multi-Robot Team Coordination n Communication – Often times relies on Master-Slave Hierarchy § Inherent brittleness to this approach – Bandwidth limitations – Robustness - Master failure? – Heterogeneous or Homogeneous approach? – Is explicit communication needed or is implicit communication enough to achieve the goal? – If communications are used how much is needed and what should be communicated?
Cooperation Without Communication n Is cooperation in a behavior based multirobot team without communication possible? n If so, how effective is it?
Behavioral Composition n Forage n Acquire n Deliver – Noise – Avoid static obstacles – Avoid robots – Move to goal – Avoid static obstacles – Noise – – Move to goal Avoid static obstacles Avoid robots Noise
Test Cases n n Simulation Map Size: 64 x 64 units Maximum Sensor Distance: 25 units Forage n Acquire/Deliver – – Noise Gain: 1. 2 Noise Persistence: 4 Avoid Obstacles Gain: 1. 0 Avoid Robots Gain: 0. 5 – – Noise Gain: 0. 2 Noise Persistence: 2 Move to Goal Gain: 1. 0 Avoid Robots Gain: 0. 1
Results n 2 Robots / 1 Attractor
Results n 4 robots / 4 attractors
Results n Without using communication, the simulation still shows coherent cooperation between the team members – Cheaper hardware – Fewer points of failure
References n Arkin, Cooperation without communication n For more quantitative comparisons between levels of communication see: Balch/Arkin, Communication in Reactive Multiagent Robotic Systems
Behavior Based Multi-Robot Team Coordination Continued n Local versus Global Control Laws – Local Control § Simple and contain emergent properties § Oftentimes unclear as to how to design local control laws – Global Control § Allow for more coherent team cooperation § Often results in increased communication requirements
Global Control Laws n Global Goal Knowledge – Information concerning the overall goal of the agents behavior – Can be encoded into robot if the goal is not dynamic n Global Knowledge – Information concerning what other robots are doing – Information concerning what other robots will do Obtaining this knowledge must often come from outside sources n The knowledge is computationally costly n Oftentimes all the needed global knowledge is not known n
Local Control Laws n Computationally Simple n Handle dynamic environments well n Oftentimes do not produce optimal results n Must rely on physical sensors
Experiment Simulation of mission involving formation maintenance while moving to goal n 4 different strategies with increasing global control n Quantitatively measured via deviation from the formation and time taken to reach goal n
Experiment Continued n Strategy I: Local Control Only – Effective for smooth trajectories – Sharp turns cause formation to break up due to local control § Robots maintain their position by staying a fixed distance from a certain side of neighbor
Experiment Continued n Strategy II: Local Control Augmented by Global Goal – Robots given knowledge of global goal: Maintain line formation – Robot D now moves to a more globally appropriate position – May be inappropriate if B is merely avoiding obstacle
Experiment Continued n Strategy III: Local Control Augmented with Global Goal and Partial Global Information – At time of robot B’s turn, other robots are informed of destination of waypoint X
Experiment Continued n Strategy IV: Local control augmented by global goal and more complete global information – Robots are given complete knowledge of leaders route – Allows other robots to predict future positions of the leader and resulting positions for themselves
Results As global information increases formation error and completion time decreases n Global goals are useful to incorporate if goals are known at run time n Global information is useful in static, well defined environments n
Results Continued Local control in situations where accomplishing the task as opposed to how the task accomplished often provides a suitable approximation of optimal behavior n Behavioral analysis in local control my approximate global knowledge n General Rule: “Local control information should n be used to ground global knowledge in the current situation. This allows the agents to remain focused on the overall goals of their group while reacting to the dynamics of their current situation. ”
References Designing Control Laws for Cooperative Agent Teams n Parker,
Questions n ? ? ?
- Slides: 31