Adaptive Dynamics Planning for Robot Navigation
Yuanjie Lu, Mingyang Mao, Linji Wang, Tong Xu, Xiaomin Lin, Xuesu Xiao
AI summary
Problem
Hierarchical navigation systems often fail in constrained environments due to a mismatch between global path planning and local dynamics execution. Existing integrated approaches rely on fixed, hand-crafted schedules to reduce dynamics fidelity, which cannot adapt to varying environmental complexity and lead to either computational waste or insufficient physical realism.
Approach
The authors introduce Adaptive Dynamics Planning (ADP), which uses a reinforcement learning agent to continuously modulate dynamics parameters like integration intervals and collision-checking resolution based on real-time environmental feedback, dynamically balancing modeling accuracy with computational efficiency.
Key results
- Integrated ADP into three classical planners and a standalone navigation system
- Achieved up to 99.61% task success rate in BARN simulations, surpassing DDP baselines
- Reduced average collision rates to 0.33% while maintaining faster navigation times
- Validated consistent performance gains across diverse simulated environments and real-world Clearpath Jackal deployments
Why it matters
Provides a robust, auto-tuning framework for real-time robot navigation that eliminates manual parameter tuning and improves reliability in complex, dynamic environments for autonomous delivery and service robots.
Abstract
Autonomous robot navigation systems often rely on hierarchical planning, where global planners compute collision-free paths without considering dynamics, and local planners enforce dynamics constraints to produce executable commands. This discontinuity in dynamics often leads to trajectory tracking failure in highly constrained environments. Recent approaches integrate dynamics within the entire plan- ning process by gradually decreasing its fidelity, e.g., increasing integration steps and reducing collision checking resolution, for real-time planning efficiency. However, they assume that the fidelity of the dynamics should decrease according to a manually designed scheme. Such static settings fail to adapt to environmental complexity variations, resulting in computational overhead in simple environments or insufficient dynamics consideration in obstacle-rich scenarios. To overcome this limitation, we propose Adaptive Dynamics Planning (ADP), a learning-augmented paradigm that uses reinforcement learning to dynamically adjust robot dynamics properties, enabling plan- ners to adapt across diverse environments. We integrate ADP into three different planners and further design a standalone ADP-based navigation system, benchmarking them against other baselines. Experiments in both simulation and real-world tests show that ADP consistently improves navigation success, safety, and efficiency.