EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

Uncertainty-Aware Adaptive Replanning: Bridging Distributional RL and Dynamic Window Approach

Xiaohan Liu, Youbing Feng

IEEE Access · 2026 · 14 · 65791-65806

DOI: 10.1109/access.2026.3688281 ↗

Kurz gefasst

Fachartikel in der Zeitschrift IEEE Access, 2026. Verfasst von Xiaohan Liu und 1 weiteren Autor:innen. Die Arbeit nennt 37 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Path planning for mobile robots in industrial environments is frequently compromised by pseudo-static obstacles with uncertain positional drift and strict kinematic constraints. Traditional discrete sampling methods, such as the Dynamic Window Approach (DWA), suffer from the curse of dimensionality and discretization artifacts, leading to control jitter and local optima. To address these challenges, this paper proposes a Distributional Risk-aware Continuous Dynamic Window (DR-CDW) framework. First, we introduce a Continuous Manifold Policy Optimization method utilizing a Differentiable Projection Operator, which embeds rigid dynamic window constraints into a continuous policy network, effectively eliminating discretization errors and ensuring $C^{1}$ control continuity. Second, to enhance learning efficiency, we employ a Physics-Informed Residual Policy architecture that superimposes learned corrections onto a baseline controller. Third, to manage environmental uncertainty, we propose a Risk-Sensitive Distributional Evaluation mechanism. This integrates Quantile Regression to model return distributions and enforces a Conditional Value at Risk (CVaR) constraint, allowing the robot to explicitly avoid long-tail collision risks. Furthermore, an Adaptive Inflation Radius strategy dynamically adjusts safety margins based on predicted uncertainty. Extensive experiments demonstrate that DR-CDW achieves a 90.15% success rate in chaotic environments and superior control smoothness compared to state-of-the-art baselines.

Zitieren

APA

Liu, X., & Feng, Y. (2026). Uncertainty-Aware Adaptive Replanning: Bridging Distributional RL and Dynamic Window Approach. IEEE Access, 14, 65791-65806. https://doi.org/10.1109/access.2026.3688281

Vancouver

Liu X, Feng Y. Uncertainty-Aware Adaptive Replanning: Bridging Distributional RL and Dynamic Window Approach. IEEE Access. 2026;14:65791-65806. doi:10.1109/access.2026.3688281

BibTeX
@article{Liu2026,
  author = {Liu, Xiaohan and Feng, Youbing},
  title = {Uncertainty-Aware Adaptive Replanning: Bridging Distributional RL and Dynamic Window Approach},
  journal = {IEEE Access},
  year = {2026},
  volume = {14},
  pages = {65791--65806},
  doi = {10.1109/access.2026.3688281}
}
Kostenlos starten →