Uncertainty-Aware Reactive Control for Signal Temporal Logic with Unknown Obstacles

In this paper, we investigate the challenge of con-trol synthesis for Signal Temporal Logic (STL) specifications within workspaces harboring unknown obstacles. We consider a sensing-based control setting, where robots are equipped with sensors with a limited sensing range to acquire obstacle locations with prior information. The objective is to synthesize a control strategy that maximize the robustness degree of the STL tasks while simultaneously minimizing the probability of collisions with obstacles during its operation. To address this, we present a receding horizon reactive control framework, tackling the problem by solving an uncertainty-aware optimal planning problem at each time instant. A key aspect of our approach is the avoidance of explicit computation of collision probability. Instead, we transform the uncertainty-aware synthesis problem into a non-probabilistic optimization problem, and the results of which still provide probabilistic guarantees. Simulation results are provided to illustrate the effectiveness of our approach.