An End-To-End Trajectory Planner for Safe and Efficient Navigation in Crowded Dynamic Environments
Shuting Zhang, Haowen Wang, Guangchen Li
AI summary
Problem
Traditional hierarchical planners degrade with unpredictable dynamic obstacles, while existing learning-based methods rely on heavy semantic detection, complex reinforcement learning, or simplified motion models that fail in cluttered spaces.
Approach
The framework processes LiDAR range images and temporal residuals through a dual-branch network to predict trajectory constraints for motion primitives, guided by a unified physics-based loss and a repulsion-based region adjustment for crowded scenarios.
Key results
- Lightweight sparse map representation for temporally continuous dynamic obstacles
- Repulsion-based planning region adjustment for highly crowded scenarios
- Millisecond-level planning latency with high safety and trajectory smoothness
- Superior success rate and flight efficiency over baselines in diverse dynamic simulations and real-world tests
Why it matters
Enables real-time, computationally efficient autonomous navigation for UAVs in complex, unpredictable environments without relying on heavy semantic processing or expert tuning.
Abstract
This paper presents a novel end-to-end trajectory planning framework that integrates LiDAR-based perception with trajectory optimization, enabling safe and efficient navi- gation in dynamic environments without relying on semantic de- tection or explicit kinematic modeling. Learning-based dynamic collision avoidance methods often depend on reinforcement learning, which introduces challenges related to training effi- ciency, model generalization, and deployment safety. To address these limitations, we propose a lightweight map representa- tion for temporally continuous dynamic obstacles, facilitating unsupervised network training with physically simulated data. Additionally, a repulsion-based adjustment method built upon motion primitives allows adaptive trajectory planning in highly crowded scenarios where no feasible trajectory exists, balanc- ing target-reaching objectives with motion safety. Extensive simulations and real-world experiments demonstrate that the proposed framework achieves millisecond-level planning latency while ensuring high safety, trajectory smoothness, and flight efficiency. The demonstration video is available on the project website: https://swift520.github.io/Dynamic-Planner/.