PUL-SLAM: Path-Uncertainty Co-Optimization with Lightweight Stagnation Detection for Efficient Robotic Exploration

Existing Active SLAM methodologies face issues such as slow exploration speed and suboptimal paths. To address these limitations, we propose a hybrid framework combining a Path-Uncertainty Co-Optimization Deep Reinforcement Learning framework and a Lightweight Stagnation Detection mechanism. The Path-Uncertainty Co-Optimization framework jointly optimizes travel distance and map uncertainty through a dual-objective reward function, balancing exploration and exploitation. The Lightweight Stagnation Detection reduces redundant exploration through Lidar Static Anomaly Detection and Map Update Stagnation Detection, terminating episodes on low expansion rates. Experimental results show that compared with the frontier-based method and RRT method, our approach shortens exploration time by up to 65% and reduces path distance by up to 42%, significantly improving exploration efficiency in complex environments while maintaining reliable map completeness. Ablation studies confirm that the collaborative mechanism accelerates training convergence. Empirical validation on a physical robotic platform demonstrates the algorithm’s practical applicability and its successful transferability from simulation to real-world environments.

💡 Research Summary

The paper “PUL-SLAM: Path-Uncertainty Co-Optimization with Lightweight Stagnation Detection for Efficient Robotic Exploration” presents a novel hybrid framework to address the long-standing challenges in Active SLAM: slow exploration speed and suboptimal, inefficient robot trajectories. The authors identify that existing methods—including traditional frontier-based approaches, sampling-based techniques like RRT, and information-theoretic methods—often optimize for a single objective (e.g., coverage or uncertainty reduction) and lack adaptive mechanisms to handle inefficient behaviors during exploration. While Deep Reinforcement Learning (DRL) offers promise, its performance is highly sensitive to reward function design, and it can still lead to local oscillations or redundant exploration.

To overcome these limitations, PUL-SLAM integrates two core innovations. The first is a Path-Uncertainty Co-Optimization DRL Framework. This framework features a dual-objective reward function that jointly optimizes for shorter travel distance and lower robot pose uncertainty (measured using the D-optimality criterion). This design explicitly balances the exploration-exploitation trade-off. A unique “exploration efficiency” metric is calculated as the ratio of newly mapped area to distance traveled. When this efficiency falls below a threshold, an additional path penalty is applied, actively discouraging the robot from engaging in unproductive movements.

The second key component is a Lightweight Stagnation Detection (LSD) mechanism. This module operates in real-time to identify and abort inefficient exploration episodes, preventing the reinforcement learning agent from learning these bad behaviors. It consists of two parts: 1) LiDAR Static Anomaly Detection, which analyzes the similarity between consecutive LiDAR scans to detect if the robot is stuck or oscillating without meaningful progress, and 2) Map Update Stagnation Detection, which monitors the rate of new area added to the map. If either detector triggers, the current exploration episode is terminated immediately and the environment is reset.

Extensive experiments were conducted in complex simulation environments and on a physical TurtleBot3 platform. The results demonstrate that PUL-SLAM significantly outperforms benchmark methods like frontier-based exploration and RRT. It achieves reductions of up to 65% in exploration time and up to 42% in total path distance, while maintaining reliable map completeness. Ablation studies confirm the critical role of both the stagnation detection mechanism and the path penalty term in accelerating training convergence and improving final performance. The successful transfer of the simulation-trained model to the real robot validates the practicality and robustness of the proposed approach, effectively bridging the sim-to-real gap. PUL-SLAM thus provides a comprehensive solution that enhances the efficiency, adaptability, and robustness of autonomous robotic exploration systems.