The University of Sussex

Plan recognition, dynamic world modelling and plan elaboration in the autodrive system for rapidly changing, uncertain, multi-agent environments

Sharon Wood

Autodrive is a planning system for rapidly changing multi-agent environments. It simulates the generation and execution of a driver's plan to reach a destination safely, taking account of other road users and obeying traffic signs and signals, in a simulated microworld. The behaviour of each vehicle in the microworld is generated by clones of the processes governing the behaviour of the driver. The system architecture is based upon a re-appraisal of a planner's representational needs for interacting with such an environment and includes components for recognising the intentions of other drivers and hypothesising their future behaviour through a process of 'dynamic world modelling'. An initially abstract plan is progressively refined until a sequence of actions are identified which satisfy the multiple goals of the driver. The selection of appropriate goals is made through a process of 'dynamic goal creation'. Plan generation and execution is interleaved in a cyclic process, progressively directing the behaviour of the driver as he undertakes his journey.

This paper is not available online