Integrating State-Based Multi-Agent Task Allocation and Physical Simulators

The multi-robot task allocation (MRTA) systems face the challenge of adapting to dynamic environments where new tasks and communication errors might appear during execution. This paper presents a framework to run agent-based MRTA within a physical simulator to test different algorithms and/or setups. Agents are modeled by a specific type of state machines able to represent deliberative behaviors as well as reactivity. While this adds formality and simplifies implementation, execution of state machines within a physical simulator requires decoupling transitions that imply the passing of time from those occurring instantly. The result framework includes a state machine execution engine that synchronizes with the simulator’s engine. Experiments using an auction-based MRTA for an example plant show not only the capability of the framework for modeling a wide range of systems but also that the MRTA method works with on-the-fly task inclusions, varying number of active robots and error occurrences.