Towards an Adaptive Robot Control Architecture

 

Robotic Urban Search and Rescue involves the location, extrication, and initial medical stabilization of victims trapped in confined spaces using mobile robots. Such rescue operations raise several issues. Part of them are studied in the AROUND project. The AROUND (Autonomous Robots for Observation of Urban Networks after Disaster) project aims at designing an automated observation system for disaster zone in developing countries like Vietnam. The idea is to deploy a large number of autonomous mobile robots able to self-organize in order to collect the information in impacted urban sites and to dynamically maintain the communication links between rescuers. Robots involved in rescue operations have to be reactive while smart enough to deal with complex situations. Hybrid agents seem to be valuable architectures for controlling such robots. A such architecture combines a fast reactive layer with a more deliberative one dealing with long term planning. However, most existing models of hybrid agents commit in early design stages to some particular software agent architecture. The resulting robots fit then only a restricted application context. They quickly become inappropriate when the execution context changes. One possible change in the execution context is the use of robots with different capabilities and resources. The same missions can be performed differently (reactively or in a more deliberative way) according to robots resources. Therefore, it is interesting to be able at deployment-time to tune agents "hybridation", i.e. switch some tasks from the reactive layer to the deliberative one or vice versa. That is adapting the agent architecture statically between two rescue operations. Robot's control architecture should not only bear static adaptation, but it should also allow dynamic adaptation. Robots controlled using such architecture has to react to the evolution of their environment (evolution of resources, robot failures, ...) in order to be as efficient as possible. In this paper we present our on-going work on adaptive hybrid agent architectures. Our approach relies on software components. Components are software entities with explicit dependencies. This property eases adaptation through components replacement and re-assembling.