Multiagent control of Self-reconfigurable Robots -Hristo bojinov,Arancha Casal,Tad Hogg

Harini Gurusamy

Features

Identical simple Modules Dynamic adaptation Multiagent Control De-Centralized Control No complex sensors

Real time e.g

Ant Colony

Amoeba

Two approaches

o Combinatorial Searcho Identify motion of modules 1.Module Specificationo Not suitable to real timeo 2.Agent based Controlo Creating structures with propertieso Decomposes control problems

Assumptions

Limited computational capabilities Limited Memory Simple FSM No global broadcast Limited communication

Experimental Robot PlatformProteo-Modular self reconfigurable robot

Polypod

Features

Substrate reconfiguration Geometric constraintso Dodecahedrao Max Internal volumeo Rotations-120 Complexity-12 face-Actuation Direct Communication

Simulations

Asynchronous operation Behaviour execution Different random order Denial of movement Notification to control programs No power limits

How a module samples from the availabe moves??

Equal probability Directed random move Positive dot product E.g.direction of ball

Control Primitives

Growth Seed Scents Mode-present FSM

Modules-search,seed,Final,Node

Recursive branching for Locomotion & Manipulation

SLEEP SEARCH SEED FINAL NODE-spawns seeds INODE-emit node scent & propagates

regular scent

Dynamic adaptation to external forces

Supporting weight on legs Neglect weights of modules 1 level and 2 level branching Fixed Module Transmit scentRoot Module IROOT-Avg wt > Fmax with Pmax AROOT ROOT-Avg wt < Fmin with Pmin

Contd..

Probabilistic approach Avoids oscillations Time order -200 cycles Probability to change state-1/R Impact on real world?????????

Grasping objects

SLEEP SEARCH SEED TOUCH & TOUCHSEED FINAL

Local minima & Stability

Physical motion Constraints Surface scent Alter the design Careful reconfiguration rules

Conclusion

Local simple rules Genetic algorithms & FPGA Protein motors

The USC/ISI CONRO Project