Multimodal Action Diffusion for Robust End-to-End Autonomous Driving
2026-06-01 • Computer Vision and Pattern Recognition
Computer Vision and Pattern Recognition
AI summaryⓘ
The authors study how autonomous driving systems decide the car's controls like steering, throttle, and braking directly, rather than first planning a path and then controlling the car. They argue that allowing the system to consider multiple possible actions, rather than just one fixed action, improves driving performance and stability. To do this, they introduce the Action Diffusion Transformer (ADT), which generates several candidate actions and picks the best one during driving. Their experiments show that ADT works better and faster than previous methods on a tough driving test, proving that thinking about multiple possible actions is important for self-driving cars.
End-to-End Autonomous DrivingTrajectory WaypointsControl Signal PredictionAction MultimodalityDiffusion TransformerMean Squared ErrorNearest Neighbour MatchingBench2Drive BenchmarkDriving RepresentationsClosed-Loop Driving
Authors
Jorge Daniel Rodríguez-Vidal, Diego Porres, Gabriel Villalonga Pineda, Antonio M. López Peña
Abstract
End-to-End Autonomous Driving (E2E-AD) systems have largely converged on predicting intermediate trajectory waypoints, delegating final control to hand-crafted controllers with GPS access. Direct control-signal prediction (outputting throttle, steer and brake in an end-to-end fashion) remains underexplored, and critically, the role of action multimodality in such systems is not well understood. We argue that moving beyond deterministic, single-action outputs is not merely a modelling choice, but a key driver of driving performance, representational quality, and training stability. To validate this, we introduce the Action Diffusion Transformer (ADT), an anchor-free diffusion transformer trained with a MSE objective that natively models the multimodal distribution of plausible driving actions. Rather than committing to a single deterministic command, ADT generates K action candidates and selects the most suitable one at inference via Nearest Neighbour Matching (NNM). Beyond strong benchmark numbers, we show that action multimodality yields measurable benefits in learned representations and behavioral consistency, effects that deterministic architectures cannot replicate. ADT surpasses previous state-of-the-art on the challenging closed-loop Bench2Drive benchmark while achieving ten times lower latency, demonstrating that expressive, multimodal action modelling is both practically efficient and conceptually essential for robust end-to-end driving.