Discrete Event Systems Based Robotic Task Modeling
Most of the existing robotic task models are not based on formal approaches, are concerned only with a small number of behaviors and are typically tailored to the task at hand. We have proposed, back to 1998 [1] a systems-theory-based task modeling approach for general robotic tasks which enables a systematic approach to modeling, analysis and design, scaling up to realistic applications, providing methods for logical verification, stochastic performance, and design from specifications, as well as execution improvement over time through learning. Our approach is based on using discrete event systems (DES) models, mainly Petri nets and finite state automata, for robot plans representation. This particular representation enables using all the available DES analysis and design tools to handle robotic task formal analysis and design.
Representing robot plans by DES
A plan is composed of several actions, running concurrently or sequentially. Action sequences, loops, subplans running in separate robots from a team, are possible examples of plans under the above definition.
In our robotic task model, theree entities are defined:
- propositions (predicates with instantiated arguments), e.g., see(object), in(robot1, room3)
- primitive actions (may have arguments, but must be also bound to some actual object), e.g., move2object, standby, catch(glass)
- events, e.g., start_standby, button_pressed, wheel_failure
The subset of events corresponding to starting a primitive action are ""controllable"". Those
corresponding to occurrences the robot(s) can not start or stop (e.g., due to the environment dynamics, ot to other agents' actions) are uncontrollable.