Safer Trajectory Planning with CBF-guided Diffusion Model for Unmanned Aerial Vehicles

arXiv cs.RO / 4/21/2026

📰 NewsDeveloper Stack & InfrastructureModels & Research

共有:

Key Points

The paper proposes AeroTrajGen, a diffusion-based framework for generating safe, agile UAV trajectories during complex aerobatic maneuvers.
It introduces control barrier function (CBF)-guided sampling in the reverse diffusion process to enforce collision-free motion by combining safety constraint gradients with the model’s score function.
The approach uses an obstacle-aware diffusion transformer with multimodal conditioning (trajectory history, obstacles, maneuver style, and goal) to produce smooth, highly agile trajectories across 14 aerobatic maneuvers.
Evaluated in simulation with multiple obstacles, CBF-guided sampling cuts collision rates by 94.7% versus unguided diffusion baselines while maintaining agility and trajectory diversity.
The authors trained the model on 2,000 expert demonstrations and open-sourced the code on GitHub for reuse.

Abstract

Safe and agile trajectory planning is essential for autonomous systems, especially during complex aerobatic maneuvers. Motivated by the recent success of diffusion models in generative tasks, this paper introduces AeroTrajGen, a novel framework for diffusion-based trajectory generation that incorporates control barrier function (CBF)-guided sampling during inference, specifically designed for unmanned aerial vehicles (UAVs). The proposed CBF-guided sampling addresses two critical challenges: (1) mitigating the inherent unpredictability and potential safety violations of diffusion models, and (2) reducing reliance on extensively safety-verified training data. During the reverse diffusion process, CBF-based guidance ensures collision-free trajectories by seamlessly integrating safety constraint gradients with the diffusion model's score function. The model features an obstacle-aware diffusion transformer architecture with multi-modal conditioning, including trajectory history, obstacles, maneuver styles, and goal, enabling the generation of smooth, highly agile trajectories across 14 distinct aerobatic maneuvers. Trained on a dataset of 2,000 expert demonstrations, AeroTrajGen is rigorously evaluated in simulation under multi-obstacle environments. Simulation results demonstrate that CBF-guided sampling reduces collision rates by 94.7% compared to unguided diffusion baselines, while preserving trajectory agility and diversity. Our code is open-sourced at https://github.com/RoboticsPolyu/CBF-DMP.