Temporal Logic Task Planning for Autonomous Systems With Active Acquisition of Information

High-level task planning is one of the central problems in autonomous systems such as unmanned ground vehicles (UGV). In this context, the agent makes decisions online to ensure the satisfaction of complex tasks under dynamic environment. In practice, control decisions are made based on the information acquired by sensors that can be turned on/off. Therefore, in the task planning problem, one needs to synthesize control and sensing decisions jointly in order to achieve the tasks. In this paper, we formulate and solve a
control-sensing co-synthesis
problem for linear temporal logic (LTL) tasks. The objective is to synthesize an active-sensing controller such that a given LTL formula accepted by a deterministic Büchi automaton can always be satisfied with a provably correct formal guarantee. To solve this problem, we propose a new approach integrating offline computations with online executions. Based on the winning regions pre-computed offline, the autonomous system can generate both control and sensing decisions online to drive the system. We show that the proposed approach is both sound and complete. The scalability and effectiveness of the proposed method are evaluated by both numerical experiments and hardware implementations in UGV task planning.