DeeAD: Dynamic Early Exit of Vision-Language Action for Efficient Autonomous Driving

TL;DR

DeeAD is a training-free early-exit framework for vision-language action models in autonomous driving, significantly reducing inference latency by evaluating trajectory feasibility and adaptively skipping redundant layers.

Contribution

DeeAD introduces a novel, training-free early-exit method with a multi-hop controller that enhances efficiency of VLA models without retraining.

Findings

Achieves up to 28% transformer-layer sparsity

Reduces latency by 29%

Maintains planning quality and safety

Abstract

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.