Robot navigation that plans with full dynamics only where it matters — 1st place, 2025 BARN Challenge (simulation)
Overview
Most navigation planners either pay the full computational cost of planning with complete robot dynamics everywhere, or ignore dynamics and suffer in difficult terrain. Decremental Dynamics Planning (DDP) (Lu et al., 2025) starts planning with full dynamics near the robot — where fidelity matters most — and progressively simplifies the dynamics model along the horizon, spending compute where it buys the most safety and performance.
Top: a conventional global + local planner. Bottom: DDP plans one trajectory with dynamics fidelity that decreases along the horizon.
My contribution
This is a collaboration led by my labmate Yuanjie Lu (with Tong Xu, Nick Hawes of Oxford, and our advisor Xuesu Xiao); I am third author and contributed to the planning framework and experimental evaluation.
Published results
Augmenting three different existing planners with DDP improved planning performance across the board.
A new DDP-based navigation system won 1st place in the simulation phase of the 2025 BARN Challenge (Benchmark Autonomous Robot Navigation), validated in both simulated and physical experiments.
Paper
Y. Lu, T. Xu, L. Wang, N. Hawes, X. Xiao. “Decremental Dynamics Planning for Robot Navigation,” IROS 2025. arXiv
Winner of the simulation phase of the 2025 BARN Challenge (Benchmark Autonomous Robot Navigation) at ICRA 2025.
Most, if not all, robot navigation systems employ a decomposed planning framework that includes global and local planning. To trade-off between efficiency and efficacy, current systems have to balance the complexity of the dynamics model and the horizon of the planner. In this paper, we present Decremental Dynamics Planning (DDP), which incorporates dynamics into the global planning process near the robot and gradually simplifies the dynamics model along the planning horizon, allowing the planner to consider full dynamics where they matter the most. We augment three different planners with DDP and observe improved planning performance overall. We also build a new DDP-based navigation system that achieves first place in the simulation phase of the 2025 BARN Challenge, validated in both simulated and physical experiments.
@inproceedings{lu2025decremental,title={Decremental Dynamics Planning for Robot Navigation},author={Lu, Yuanjie and Xu, Tong and Wang, Linji and Hawes, Nick and Xiao, Xuesu},booktitle={2025 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS)},year={2025},}
