Mode-Switching Frontier Search for Robust Autonomous Exploration in Complex Terrain

Autonomous exploration in unstructured environments is often hindered when robots become trapped in navigational deadlocks, causing mission failure. Frontier-based planners, in particular, often lack a strategic mechanism to recover when all immediate paths are blocked. This limitation often leads to a state of `navigation at a dead end’, where an autonomous agent without a recovery strategy can become permanently stuck. The framework to detect such impasses, leverage memory of past states, and strategically recover is therefore a critical and often missing component for robust autonomous agents. To address a critical failure mode in frontier-based exploration where robots become trapped in navigational deadlocks, we introduce a state-aware supervisory architecture. This framework enhances the resilience of standard local planners by integrating a robust recovery mechanism that systematically switches the robot between Search, Recovery, and Exploration modes. Upon detecting a stuck state by monitoring its trajectory history, the system enters a dedicated recovery mode. In this mode, it consults a global database of previously discovered frontiers—acting as a short-term memory—to select an optimal recovery point and escape the impasse. We validated our framework in challenging simulation scenarios where a standard local planner fails. Experimental results show that our framework transforms a local planner with a 0% success rate into a resilient system achieving up to a 100% success rate, demonstrating its effectiveness as a critical component for robust autonomous navigation.