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

TL;DR

The paper tackles slow exploration and suboptimal trajectories in Active SLAM by introducing PUL-SLAM, a dual-layer framework that couples Path-Uncertainty Co-Optimization DRL with Lightweight Stagnation Detection. The high-level DRL optimizes travel distance and pose uncertainty through a dual-objective reward, while LSD identifies motion failures and stagnation to terminate ineffective episodes, improving training stability. Empirical results in simulation show substantial gains in exploration speed and path efficiency over frontier-based and RRT baselines, with ablations confirming the complementary benefits of PUR and LSD; real-world experiments further demonstrate sim-to-real transferability. The work advances autonomous exploration by delivering a robust, efficient, and transferable Active SLAM solution suitable for complex environments and practical deployment.

Abstract

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.

Figures (12)

• Figure 1: The overall framework of the PUL-SLAM system.
• Figure 2: For the sampling of LiDAR data, the data volume changes from 360 to $N$.
• Figure 3: Env-1 for training.
• Figure 4: Trajectory and mapping results in Env-2.
• Figure 5: Trajectory and mapping results in Env-3.