Heterogeneous agents for multi-robot cooperation

Abstract

A multi-agent architecture, comprising one reflective agent allocating tasks among a number of behavioural agents, has been shown to demonstrate several advantages. Much of the detailed reasoning involved in achieving a goal undertaken by predictive control agents is not required by our reflective agent owing to the behavioural agents' capabilities. This means the reflective agent only requires an abstract model of the environment, typically containing information only of static features within the environment. Much less work must be done to update the world model. Behaviours have been designed that enable the behavioural agents to cooperate by `non-symbolic' communication. Research is currently focussing upon the design of more cooperating behaviours such as a herding behaviour and a formation behaviour. It is hoped to implement these using nonlinear sensing. The architecture is flexible and allows for behavioural agents of differing capabilities to cooperate within the environment. The use of agent models as the interface to the actual agents allows the reflective agent to communicate with behavioural agents of differing architectures. (7 pages)