NobleBlocks

Research indexConference paper

Robustness quantitative evaluation and algorithm optimization for path planning in unknown scenarios

Liu

International Conference on Computer Vision, Algorithm, and Image Processing (CVAIP 2026)19 Oct 2026

DOI10.1117/12.3134860

Paywalled

The full text is at the publisher

You can read the abstract here, or open it at the publisher.

Read at the publisher

Abstract

Path planning in unknown scenarios is challenging because the planner must cope with partial observability, sensor noise, dynamic obstacles, and terrain uncertainty. Classical grid or sampling based planners produce interpretable trajectories but degrade rapidly when the environment differs from the map, while end to end deep reinforcement learning policies react to disturbances but often fail to generalize beyond training distributions. This paper studies how to jointly quantify and improve the robustness of path planners in unknown scenes. We first propose a composite robustness metric R that unifies success rate, path efficiency, and safety margin across a batch of perturbed trials, so that a single scalar reflects planner behavior under distributional shift. Building on this metric, we introduce the Hybrid Robust Route Planner (HRRP), which couples a D* Lite global planner with a PPO local policy that shares a convolutional occupancy encoder and is trained with domain randomization. Extensive experiments on four unknown scene benchmarks show that HRRP raises the composite robustness score from 0.69 to 0.84 while improving success rate by 11 percentage points and reducing path length variance by 42 percent relative to the strongest baseline. Ablation results confirm that both the hybrid coupling and the randomization schedule are necessary for the observed robustness gain.

Discussion

More

Suggest a correction or report