Distributed and Reactive Controller Synthesis for Multi-Agent Systems Under Finite Horizon Temporal Logic Tasks

Automated synthesis of local controllers for multi-agent systems to satisfy complex task specifications has attracted extensive attention. However, it remains unclear how to formally guarantee that their composed behavior satisfies the specified global task. In this paper, we aim to synthesize an automated and distributed control strategy for fleet-wise tasks specified as linear temporal logic formulas, such that the agents act asynchronously and synchronize only on shared actions via local coordination. The proposed method consists of three main steps. First, the set of satisfying global control strategies is computed via parallel composition. Among these strategies the conditions for decomposability are evaluated, based on which the global strategy over a maximum synchronization scheme is found. Then the synchronization scheme is further refined to obtain more efficient local control strategies. It is formally proven that the resulting global behavior along with the synchronization scheme satisfies the specified global task. Also, the synthesized local controllers are reactive to changes in the workspace or in the fleet during online execution, without the need for replanning. Thus, collaborative relations among the agents are adaptive as needed. Numerical simulations and hardware experiments are conducted for nontrivial scenarios. Note to Practitioners—This paper was motivated by the problem of coordinating a fleet of autonomous robots in collaborative search and delivery processes, where local controllers are synthesized for each robot such that the specified global task specification is satisfied. Existing approaches to address such problems often rely on a fully-connected communication topology and fixed collaborative relations among the robots, thus limiting efficiency of the multi-robot execution and yielding difficulty of coordination. Also, most of existing approaches cannot directly deal with the changes in the workspace or in the fleet during task execution unless a task replanning is performed, which leads to longer time of system-wide communication and task completion, yielding failures during online execution. In this paper, we propose a novel distributed control architecture to tackle these issues, where the robots are controlled to operate asynchronously and achieve online local coordination by synchronization on collaborative actions. Moreover, compared with the common solutions, the collaborative relations among the robots are formed and removed dynamically as needed and the robots are robust to uncertainty during task execution. It is formally proven that the resulted global behavior of the robot fleet along with the synchronization scheme is consistent with the specified global task. We have shown that it is particularly useful for complex and coupled multi-robot applications, where the inter-robot collaborations are feasible and local. Experimental results suggest that this approach is applicable to multi-robot systems which greatly improves the concurrency and efficiency of task execution. In the future research, we will draw inspiration from decentralized approaches for task decomposition to alleviate the computational burden as the number of robots increases.