CAREER: Linguistic Control of Mobile Robots

Georgia Tech Research Corporation, Atlanta GA

Investigators

Abstract

When humans instruct each other h w to carry out particular tasks,nly a limited number of tokenized (linguistic)instructions are used.In contrast to this,classic control theory speci .es a control action to be carried out at each time instant.But,in a number f applications,such as semi-autonomous service robots for industrial and domestic use,intelligent appliances,and communication constrained embedded and/or teleoperated devices,the control procedures have a natural,linguistic .avor.The interpretation here is that linguistic control instructions specify particular modes f peration rather than explicit control values. The verall objective f this CAREER-prop sal is t develop and disseminate rational meth- ds for understanding how continuous devices,such as mobile rob ts,should be controlled using computer generated,linguistic inputs.In particular,it will be studied how these instructions should be de .ned,selected,and coded in rder to minimize the number of bits transmitted from the computer t the rob t,while guaranteeing that the system meets its speci .cations.F r this, an information theoretic approach t control theory will be developed,serving as a useful tool not nly for source coding of control signals,but also for describing how symb lic instructions should be interpreted and perated n by the continuous systems.Questions concerning what sensors and actuators to use in a given rob tic application can be addressed quite elegantly within this framew rk as well. The development of this program will serve as a foundation for a teaching and research career in electrical and computer engineering,with particular emphasis n control theory and rob tics. In particular,the e .ort will be focused n the following key areas: Hybrid Models: Control theoretic models will be developed for bridging the gap between the continuous (the device dynamics)and the discrete (the linguistic control inputs).The proposed models will give the symbolic inputs a clear,control theoretic interpretation,at the same time as the expressive p wers of di .erent sets f control inputs can be made explicit. Complexity Measures and Coding Theory: A complexity measure,corresp nding to the number of bits needed for c ding a particular control procedure,will be proposed.Furthermore, by establishing probability distributions ver the set f control modes,optimal c ding algorithms can be derived. Multi-Modal and Multi-Agent Control Design: Strings of appropriate control modes will be constructed that make the systems meet given speci .cations while keeping the descriptions short.F r this we will design control laws that use the available sensory data in an e .cient way.These results will be extended t multi-agent control where formation maintenance and obstacle avoidance should be achieved while minimizing the inter-agent communications. Educational Plan: Due t the multidisciplinary nature f the proposed w rk,the under- graduate control curriculum will be modi .ed and tw senior level undergraduate courses will be included in the curriculum at the Georgia Institute of Technology.A multi-agent rob tic testbed where b th human and computer generated inputs can be used as a basis for the motion control will furthermore be developed. 1