Multiple Robot Systems:Task Distribution, Coordination and Localization

Sameer Singh83 ECE 2000

Final YearNSIT

Problems with Single Robots

Application Specific – Thus cannot be used for different tasks

Expensive Features – Require large amount of sensors

Fault IntolerantHeavy computation power required

MRS Advantages

Not Application SpecificEasy on the controllerHigh Fault ToleranceComes out cheaper for complex tasksCan carry out Tightly Coupled Tasks

Mobile Robot for MRS

Pioneer 3DX Robot

from ActivMedia Robots

Specs of Pioneer 3DX

250 Watt hours of hot swappable batteries8 forward sonar ringPayload up to 23 kgsSpeeds up to 1.6 m/sDimensions: 44cm X 38cm X 22cm

Internal Behaviors

Drive – controlled by keys or joystickPlan PathsDisplay map of sonarCommunicate Data – Sensor and ControlRun C/C++ ProgramSimulation Software

Accessories for Pioneer 3DX

Wireless EthernetLaser MappingRobotic ArmPan-Tilt-Zoom color CameraStereo Range-Finding CameraCompasses and Tilt SensorsInfra Red Ranging SensorsBumpers and Grippers

Task Distribution

Distribution of Tasks to different robotsEnsure Fault-free operation in dynamic,

noisy environmentEnsure CooperationSuitable for Heterogeneous EnvironmentNon-deterministic Polynomial complete

problem

Task Distribution

Tasks are divided into sub-tasks, which further into sub-subtasks, and so on… hence, a task tree is generated.

Responsibility of Parent to allocate child tasks to other robots and ensure completion

Uses broadcast communicationAuction based Task distribution

Steps for Task Distribution

1. Task Announcement

2. Metric Evaluation

3. Bid Submission

4. Auction Close

5. Progress/Contract Renewal

Resources required includes features (camera, gripper, etc.) and/or capabilities (mobile, carry heavy, open doors, etc.) and/or current state (idle, charging, pushing box, etc.)

Task Distribution

Tightly Coupled Task Watcher with ranging sensors and camera Pusher with grippers and camera

Localization

Absolute and Relative LocalizationGPS or some model-based to be used for

Absolute, thus expensiveAbsolute better for single robotsRelative more pertinent for MRS since

formations require relative, not absoluteCan develop absolute coordinates from

relative

Localization

Each robot is the originContains probability distribution of all

robotsAll robots broadcast their observationsDistributions are updated according to the

observations from self and other robotsThus robots even know where other robots

are which have not even been seen

Spatial Interference

Basic avoidance method fails

Conflict Resolution should take place

In nature, there is aggression, we shall emulate it

Spatial Interference

Each robot has a fear threshold, which decides how close it shall let the other robot come.

Robot is brave or afraid depending on if the other robot is outside or inside the fear threshold.

Fear threshold – Personal Space, Previous Wins, Fixed Hierachy, Priority

Spatial Interference

Other issues

Performance v/s

Population Dynamic

Population? Anonymity or

Identities Communication

References

Vaughan, R.T., Stoy, K., Sukhatme, G.S. and Mataric, M.J. - Go ahead, make my day: Robot Conflict Resolution by Aggressive Competition

Howard, A., Mataric, M.J. and Sukhatme, G.S. – Cooperative Relative Localization for Mobile Robot Teams: An Egocentric Approach

Batalin, M.A. and Sukhatme, G.S. – Spreading Out: A Local Approach to Multi-Robot Coverage

Gerkey, B.P. and Mataric, M.J. – Sold!: Auction Methods for Multirobot Coordination

Mataric, M.J. – Coordination and Learning in Multirobot Systems

ActiveMedia Robotics – www.activrobots.com

Thank You

I would like to mention the USC Robotics Research Lab for the source of the inspiration and the source of these references. Most diagrams and figures I have used are from their publications.