Driving · 2026-10-09

BridgeGuard: Explicit Safety Drift for Diffusion-based Autonomous Driving

Zhenjun Qiu, Jianing Huang, Dongang Liu, Baiyu Du, Yixun Niu, Hao Yang, Xinyu Huang, Chuan Hu, Shu Liu

AI-assisted summary based on the paper. Consult the linked source for methods and limitations. Benchmark numbers appear after editor verification.

TL;DR

BridgeGuard is a safety-constrained diffusion planning method for autonomous driving. It progressively strengthens a constraint during denoising to move intermediate trajectories toward a scene-dependent safety domain. It learns a time-dependent obstacle distance field (DistanceFieldNet) from BEV features using value and spatial-gradient supervision beyond expert trajectories, and injects the learned safety correction via low-dimensional curve updates while freezing the pretrained perception backbone and planner.

Why it matters

Diffusion-based planners can produce unsafe trajectories under distribution shift. BridgeGuard adds an explicit safety objective by learning a differentiable, time-dependent distance field and using it to refine intermediate denoising outputs in a curve space, aiming for cross-model generalization across diffusion planners. The paper also provides sufficient conditions for terminal safety in an idealized continuous-time bridge.

Method

Adapt Manually Bridged Models to driving by defining a scene-dependent safety domain and an Ω-distance, then progressively increasing a constraint term during denoising to form an Ω-bridge under stated assumptions.
Learn DistanceFieldNet to predict a time-dependent distance field from BEV features; train it with value and spatial-gradient supervision using queries sampled beyond expert trajectories to cover safe and unsafe regions.
Inject safety at inference by fitting denoised waypoint sequences into a low-dimensional curve space, evaluating field gradients at physical query horizons, and updating curve parameters while keeping the pretrained perception backbone and base planner frozen.

Limitation

Their terminal-safety guarantee applies to an idealized continuous-time manual bridge and does not directly extend to the learned field and discretized corrections, and the curve representation does not certify kinematic feasibility or collision freedom.

Abstract (from arXiv)

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.

Publication record

Bibliographic evidence describes where and when a record appeared. It does not validate scientific claims.

First public release: 2026-10-08 · Indexed by Pub-AI: 2026-10-09 UTC

preprint