Decentralized multi-agent control from local LTL specifications

Abstract

We propose a methodology for decentralized multi-agent control from Linear Temporal Logic (LTL) specifications. Each agent receives an independent specification to formally synthesize its own hybrid controller. Mutual satisfiability is not a priori guaranteed. Due to limited communication, the agents utilize meeting events to exchange their controller automata and verify satisfiability through model checking. Local interaction only when common atomic propositions exist reduces the overall computational cost, facilitating scalability. Provably correct collision avoidance and convergence is ensured by Decentralized Multi-Agent Navigation Functions.

The Multi-Agent Navigation Transformation: Tuning-Free Multi-Robot Navigation

“…Due to the closed-form solution, the NF framework enable applications on systems with limited computational capabilities. The NF framework has been successfully extended to handle multiple robots in centralized [15] and decentralized setups [3,16], with non-holonomic constraint [14] and has also been augmented with spacio-temporal scheduling and planning capabilities [6,4].…”

Communication-Free Multi-Agent Control Under Local Temporal Tasks and Relative-Distance Constraints

“…These tasks can be fully independent [5], [9] or partially dependent, involving requests for collaboration with the others [28], [29]. A major issue here is the decentralization of planning and control procedures.…”

“…Tasks are assigned to the agents locally over (possibly collaborative) services provided in pre-defined regions of interests visited along the agent's trajectories. The main contributions are: (i) different from [5] and [10] where a satisfying discrete plan is enough, the proposed initial plan synthesis algorithm here minimizes a cost of a satisfying plan; (ii) the proposed distributed motion controller guarantees almost global convergence and the satisfaction of relative-distance constraints, for an arbitrary number of leaders with different local goals. Most related literature only allows for a single leader [10], [26] or multiple leaders with the same global goal as the team [24], [32]; (iii) three different local coordination policies are proposed for different types of local tasks.…”

“…In particular, task specifications are expressed as LTL formulas for a single dynamical system in [4] and an automated framework is proposed to translate the task directly into a hybrid controller, which drives the system to fulfill this task. For multi-agent systems, LTL formulas have been used to specify complex high-level global tasks [2], [15], [17], [21], [22], [28], [30], local tasks [5], [9], [29] and even communication protocols [23] among the agents.…”

A receding horizon approach to multi-agent planning from local LTL specifications

“…Fragments of LTL have been considered for vehicle routing problems for unmanned aerial vehicles in [11], and for search and rescue missions in [20]. A decentralized control of a robotic team from local LTL specification with communication constraints is proposed in [7]. However, the specifications there are truly local and the agents do not impose any requirements on the other agents' behavior.…”