Sun, Hengyang (Tongji University), Li, Meng (Tongji University), Nie, Jialei (Tongji University), Min, Jing (Tongji University), Huang, Yanjun (Tongji University)

Planning-Oriented Trajectory Generation with Motion Prior Guidance Using Truncated Diffusion

Scheduled for presentation during the Regular Session "S26a-Motion Planning, Trajectory Optimization, and Control for Autonomous Vehicles" (TH-LM-T26), Thursday, November 20, 2025, 11:30−11:50, Broadbeach 1&2

2025 IEEE 28th International Conference on Intelligent Transportation Systems (ITSC), November 18-21, 2025, Gold Coast, Australia

This information is tentative and subject to change. Compiled on October 18, 2025

Abstract

Trajectory generation plays a crucial role in autonomous driving, requiring the production of feasible future paths aligned with human-like behaviors. While recent studies have adopted diffusion models for this task, they face two key challenges: (1) the absence of planning-aware motion priors may result in suboptimal generations, limiting behavior plausibility and alignment with human intent, and (2) high computational overhead due to lengthy iterative denoising. To address these challenges, this paper proposes a planning-oriented trajectory generation framework that integrates motion prior guidance with a robustness-enhanced truncated diffusion process. Our architecture first employs a Transformer encoder to extract and fuse spatiotemporal scenario features, coupled with a motion predictor to infer agents' future behavioral intentions. Based on these predicted modes, the model samples a corresponding trajectory anchor with noise as coarse guidance, which is then refined through a limited number of diffusion steps. Experiments on the Waymo Motion Dataset show that our method achieves competitive performance compared to prior baselines. Furthermore, the proposed framework enables robust and controllable trajectory generation under various behavioral predictions, while significantly reducing inference process down to only 5 denoising steps—thereby offering an effective balance between performance, efficiency, and flexibility.