No-regret path planning for temporal logic tasks in partially-known environments

Abstract

In this paper, we investigate the graph-based robot path planning problem for high-level specifications described by co-safe linear temporal logic (scLTL) formulae. Our focus is on scenarios where the map geometry of the workspace is only partially-known. Specifically, we assume the existence of unknown regions, where the robot lacks prior knowledge of their successor regions unless it physically reaches these areas. In contrast to the standard non-deterministic synthesis approach that optimizes the worst-case cost, in the paper, we propose using regret as the metric for planning in such partially-known environments. Regret measures the difference between the actual cost incurred and the best-response cost the robot could have achieved if it were aware of the actual environment from the start. We present a formal model for this problem setting and develop an efficient algorithm to find an optimal strategy in the sense that it meets the scLTL specification while minimizing the regret of the strategy. Our approach provides a quantitative method for evaluating the trade-off between exploration and non-exploration, rather than relying on the heuristic determinations used in many existing works. Case studies on firefighting and collaborative robots are provided to illustrate the effectiveness of our framework. Furthermore, we conduct numerical experiments on a large number of randomly generated systems and compare the performance of the regret-based strategy with other path planning strategies. The experimental results indicate that regret is a highly meaningful metric for path planning in partially-unknown environments, especially in cases where no probabilistic a priori knowledge is available.