Answers are grounded in this paper and cite their sources.
Unmanned ground vehicles are indispensable for autonomous navigation in unstructured environments, particularly in critical applications such as search and rescue, disaster response, and exploration. However, the inherent challenges of dynamic terrain conditions and uncertain traversability complicate motion planning in such scenarios. To address these challenges, this paper introduces a probabilistic traversability risk-aware one-stage motion planning framework, which integrates three key innovations: a deep learning-based perception system augmented with Bayesian regression to probabilistically assess terrain traversability and quantify risk; an entropy-based risk measure that consolidates probabilistic traversability estimates into a unified decision-making metric for robust motion planning; and an accelerated model predictive path integral control strategy, optimized via Numba-based parallelization, to ensure globally optimal yet locally feasible trajectories in real time. Extensive BenchNav simulations validate the framework’s superior performance, exhibiting the highest success rate while achieving a 4.5% reduction in travel time compared to conventional two-stage planners (e.g., A*-DWA) and generating 24.7% shorter paths than existing one-stage methods (e.g., CL-RRT). Crucially, the probabilistic traversability evaluation approach proves more robust than mean estimation in uncertain, high-risk unstructured scenarios. Furthermore, real-world vehicle experiments validate the method’s real-time computational feasibility and effective risk-aware navigation in unstructured terrains.
No takes yet. Share an insight, caveat, or question.
Fan et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: