A Data-Driven Model for Predicting Fault-Tolerant Safe Navigation in Multi-Robot Systems

Abstract

Earlier methods focused on reducing the forecast uncertainty for individual agents and avoiding this unduly cautious behavior by either employing more experienced models or heuristically restricting the predictive covariance. Findings indicate neither the individual prediction nor the forecast uncertainty have a major impact on the frozen robot problem. The result is that dynamic agents can solve the frozen robot problem by employing joint collision avoidance and clear the way for each other to build feasible pathways. Potential paths for safety evaluation are ranked according to the likelihood of collisions with known objects and those that happen outside the planning horizon. The whole collision probability is examined. Monte Carlo sampling is utilized to approximate the collision probabilities. Designing and selecting routes to reach the intended location, this approach aims to provide a navigation framework that reduces the likelihood of collisions.