G2DP: Diffusion Planning with Spatio-Temporal Grid Guidance

2026-06-24

AI Summary

Diffusion-based planning struggles in densely interactive scenarios due to inherent stochasticity, often failing to simultaneously ensure trajectory safety and route adherence, primarily owing to the absence of effective guidance mechanisms. This work proposes G2DP, the first method to embed a dense spatiotemporal grid as a differentiable guidance signal directly into the diffusion inference process. By fusing future occupancy probabilities with a route progress map, G2DP constructs a spatiotemporal cost volume modeled as a continuous safety energy functional, which injects dense gradients during denoising to steer the generation of safe and efficient trajectories. Evaluated on nuPlan, G2DP outperforms the strongest imitation learning baseline by 7.2 points and maintains superior performance under zero-shot transfer to interPlan and DeepScenario, achieving a 10.15% improvement in collision avoidance.

Abstract

In autonomous driving, diffusion-based planners have emerged as a promising paradigm for robust motion planning in dense and interactive traffic, as they can effectively model diverse driving behaviors. However, their inherent stochasticity often requires explicit guidance during denoising to ensure safety and route adherence for robust closed-loop execution. Existing guidance typically relies on sparse, entity-centric geometric queries or post-hoc refinement, yielding limited situational awareness and fragile performance in interactive scenes. To address this issue, we propose G2DP (Grid-Guided Diffusion Planning), a diffusion-based planner that directly enforces dense environmental constraints through inference-time guidance. Specifically, G2DP constructs a differentiable spatio-temporal cost volume by fusing probabilistic future occupancy distributions with a route-progress map. By formulating this volume as a continuous safety energy functional, it injects dense gradients directly into the denoising loop, actively steering trajectory generation toward collision-free and progress-optimal regions. Extensive closed-loop evaluations show that G2DP achieves state-of-the-art performance on nuPlan, outperforming the strongest imitation-learning baseline by +7.2 points in reactive score. It further maintains top scores in zero-shot transfers to interPlan and DeepScenario benchmarks, with collision avoidance improving by +10.15 over the unguided approach on interPlan. These results demonstrate that spatio-temporal cost grids serve as an effective representation for robust guidance in diffusion-based planning.

Problem

• diffusion planning
• autonomous driving
• safety guidance
• spatio-temporal constraints
• motion planning

Innovation

• diffusion planning
• spatio-temporal grid guidance
• cost volume
• inference-time guidance
• autonomous driving