Petri Net Robotic Task Plan Representation: Modelling, Analysis and Execution

Abstract

As the usage of robots in everyday tasks increases, there is a need to improve our knowledge concerning the execution of those robotic tasks. Robotic task models are usually not based on formal approaches but tailored to the task at hand. Applying discrete event system concepts to model robotic tasks provides a systematic approach to modelling, analysis and design, scaling up to realistic applications, and enabling analysis of formal properties, as well as design from specifications. Most of the work found on the literature concerning the design of robotic tasks using Discrete Event Systems is based on Finite State Automata for code generation (Dominguez-Brito et al., 2000), qualitative specifications (Kosecka et al., 1997), some quantitative specifications (Espiau et al., 1995), modularisation (Kosecka et al., 1997) and even to model multi-robot systems (Damas & Lima, 2004). Work using Petri nets to design robotic tasks under temporal requirements, focusing also on the generation of real-time, error-free code can be found in (Montano et al., 2000). Petri net Plans were introduced in (Ziparo & Iocchi, 2006) for design and execution of task plans. However, these do not close the loop, i.e., do not consider the actual implications of the actions on the environment, focusing mostly on the design. In this chapter we describe a Petri net based framework which allows a systematic approach for modelling, analysis and execution of robotic tasks. This framework is divided in three layers: task plan models, action models and environment models. The models range from the robot decision-making algorithms (task plan models) to the environment dynamics, due to physics and/or actions of other agents (environment models). In the proposedmodels, Petri net places represent tasks, primitive actions and logic predicates set by sensor readings. These logic predicates provide and abstraction of the world relevant features. By composing these models, and applying analysis techniques, important a priori information can be obtained regarding the properties of the task. The models are based on Marked Ordinary Petri Nets and Generalised Stochastic Petri Nets (Murata, 1989), allowing