让自动驾驶模型提前退出推理，提速近三成且不丢安全
DeeAD: Dynamic Early Exit of Vision-Language Action for Efficient Autonomous Driving
- 用轻量规划结果判断是否可提前终止推理
- 实测减少29%延迟，最多跳过28%模型层
- 无需重训练，适合部署在现有自动驾驶系统
视觉-语言-动作（VLA）模型融合感知、推理与轨迹生成，但因深层Transformer结构导致推理延迟高。我们提出DeeAD，一种无需重训练、基于动作引导的动态早退框架，通过评估中间轨迹的物理可行性来加速决策。不依赖置信度分数，当预测轨迹与轻量级规划先验（如Navigation或低精度规划）偏差小于2米时即终止推理。为提升效率，引入多跳控制器，根据分数变化率自适应跳过冗余层。DeeAD可无缝集成至现有VLA模型（如ORION），在Bench2Drive基准上实现最高28%的Transformer层稀疏度和29%的延迟降低，同时保持规划质量与安全性。
原文摘要 · Abstract (English)
Vision-Language Action (VLA) models unify perception, reasoning, and trajectory generation for autonomous driving, but suffer from significant inference latency due to deep transformer stacks. We present DeeAD, a training-free, action-guided early-exit framework that accelerates VLA planning by evaluating the physical feasibility of intermediate trajectories. Instead of relying on confidence scores, DeeAD terminates inference when predicted trajectories align with lightweight planning priors (e.g., Navigation or Low-precision Planning) within a tolerable deviation (<2m). To improve efficiency, we introduce a multi-hop controller that adaptively skips redundant layers based on the change rate of scores. DeeAD integrates into existing VLA models, such as ORION, without requiring retraining. Experiments on the Bench2Drive benchmark demonstrate up to 28% transformer-layer sparsity and 29% latency reduction, while preserving planning quality and safety.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。