Diffusion Forcing Planner: History-Annealed Planning with Time-Dependent Guidance for Autonomous Driving

2026-06-09

AI Summary

Existing learning-based motion planners struggle to balance stability and environmental adaptability due to temporal inconsistencies and mechanical replication of historical trajectories. This work proposes a diffusion-based planning framework that partitions trajectories into past, current, and future segments and introduces a heterogeneous joint denoising process. By incorporating a history annealing mechanism and a time-dependent classifier-free guidance strategy, the method dynamically integrates historical information during inference rather than merely replaying it. Evaluated in the nuPlan closed-loop benchmark, the approach significantly enhances trajectory continuity, stability, and controllability, achieving state-of-the-art performance in complex driving scenarios.

Abstract

Learning-based motion planners, despite recent progress, often suffer from temporal inconsistency. Small perturbations across frames can accumulate into unstable trajectories, degrading comfort and safety in closed-loop driving. Several methods attempt to inject history as a static conditioning signal to stabilize outputs, only to induce the planner to copy historical patterns instead of adapting to environment contexts. To address this limitation, we propose Diffusion Forcing Planner (DFP), a diffusion-based planning framework driven by history-guided control. Specifically, DFP decomposes the full trajectory into history, current and future segments, and assign independent noise levels to each segment. The model jointly denoises the historical and the future segments, enforcing a heterogeneous joint diffusion process. At inference, classifier-free guidance (CFG) is applied to steer future sampling using annealed history in a controllable manner. Closed-loop evaluation and comprehensive ablations on nuPlan show that DFP achieves competitive performance while producing continuous, stable, and controllable motion plans in complex driving scenarios.

Problem

• temporal inconsistency
• motion planning
• autonomous driving
• history conditioning
• trajectory stability

Innovation

• diffusion-based planning
• history-annealed guidance
• temporal consistency
• classifier-free guidance
• autonomous driving