Temporally Decoupled Diffusion Planning for Autonomous Driving

2026-03-26

AI Summary

Autonomous driving in dynamic urban environments requires balancing immediate safety with long-term objectives, yet existing diffusion-based trajectory prediction models treat trajectories as monolithic sequences, neglecting the temporal heterogeneity between near-term and far-term planning. To address this, this work proposes the Temporal Decoupled Diffusion Model (TDDM), which innovatively interprets noise as an indicator of missing information and introduces a “noise-as-mask” mechanism to segment trajectories and assign distinct noise levels to different temporal segments, thereby explicitly decoupling short- and long-horizon planning. We further design a Temporal Decoupled Adaptive Layer Normalization (TD-AdaLN) module and an asymmetric temporal classifier-free guidance strategy that leverages far-horizon context to guide the reconstruction of near-term states. Evaluated on the nuPlan benchmark, TDDM achieves or surpasses state-of-the-art performance, with particularly strong results on the challenging Test14-hard subset.

Technology Category

• Planning, Routing, and Scheduling: Planning with Language Models
• Computer Vision: Diffusion Models for Vision
• Humans and AI: Human-Aware Planning and Behavior Prediction

Application Category

• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

Motion planning in dynamic urban environments requires balancing immediate safety with long-term goals. While diffusion models effectively capture multi-modal decision-making, existing approaches treat trajectories as monolithic entities, overlooking heterogeneous temporal dependencies where near-term plans are constrained by instantaneous dynamics and far-term plans by navigational goals. To address this, we propose Temporally Decoupled Diffusion Model (TDDM), which reformulates trajectory generation via a noise-as-mask paradigm. By partitioning trajectories into segments with independent noise levels, we implicitly treat high noise as information voids and weak noise as contextual cues. This compels the model to reconstruct corrupted near-term states by leveraging internal correlations with better-preserved temporal contexts. Architecturally, we introduce a Temporally Decoupled Adaptive Layer Normalization (TD-AdaLN) to inject segment-specific timesteps. During inference, our Asymmetric Temporal Classifier-Free Guidance utilizes weakly noised far-term priors to guide immediate path generation. Evaluations on the nuPlan benchmark show TDDM approaches or exceeds state-of-the-art baselines, particularly excelling in the challenging Test14-hard subset.

Problem

• autonomous driving
• motion planning
• temporal dependencies
• diffusion models
• trajectory generation

Innovation

• temporally decoupled diffusion
• noise-as-mask
• asymmetric temporal guidance
• motion planning
• autonomous driving