Problem
Research questions and friction points this paper is trying to address.
Diffusion-based driving planners
Unsafe trajectories
Distribution shift
Autonomous driving
BridgeGuard: Explicit Safety Drift for Diffusion-based Autonomous Driving
🤖 AI Summary
This study addresses the vulnerability of diffusion planners to generating unsafe trajectories under distribution shifts by proposing a diffusion planning method with progressively reinforced constraints. The approach performs corrections in a low-dimensional curve space to ensure geometric coherence, introduces a scene-dependent safety region drift mechanism, and provides sufficient condition proofs for continuous-time bridging terminal safety. Furthermore, a DistanceFieldNet is designed to predict distance fields using combined value and gradient supervision, enabling safety injection through a frozen pretrained backbone. Evaluated on the Bench2Drive benchmark, the proposed method significantly improves driving scores and success rates, demonstrating strong cross-model generalization capabilities and enhanced safety performance.
📝 Abstract
Diffusion-based driving planners capture diverse behaviors but can generate unsafe trajectories under distribution shift. We propose BridgeGuard, a safety-constrained diffusion planning method that progressively strengthens a constraint term during denoising to drive intermediate trajectories toward a scene-dependent safety domain. Corrections operate in a low-dimensional curve space, promoting geometric coherence. A learned module, DistanceFieldNet, predicts a time-dependent distance field from bird's-eye-view features. Value and spatial-gradient supervision at queries sampled beyond expert trajectories teaches this field about both safe and unsafe regions. The learned field supplies the constraint term through safety injection while the pretrained perception backbone and planner remain frozen. We further establish sufficient conditions for terminal safety in an idealized continuous-time bridge. On Bench2Drive, BridgeGuard improves driving score/success rate from 87.99/74.99% to 90.88/76.36% for BridgeDrive and from 80.79/58.18% to 90.46/74.09% for $\text{DiffusionDrive}^{\text{geo}}$, demonstrating cross-model generalization.
Innovation
Methods, ideas, or system contributions that make the work stand out.
Diffusion-based Planning
Safety Constraint
Distance Field
Autonomous Driving
Distribution Shift
🔎 Similar Papers
No similar papers found.
