Paper 2604.18486
Xiaomi OneVL: One-Step Latent Reasoning and Planning with Vision-Language Explanation
- Published
- Apr 2026
- Research lab
- Independent
- Citations
- 3
- GitHub
- 461 stars
01 In brief
Summary
OneVL is a unified Vision-Language-Action (VLA) and World Model framework for autonomous driving that performs one-step latent reasoning and planning with vision-language explanations.
It addresses the latency of explicit Chain-of-Thought (CoT) reasoning by compressing reasoning into compact latent tokens supervised by dual auxiliary decoders: a language decoder that reconstructs text CoT and a visual world model decoder that predicts future-frame tokens, forcing the latent space to internalize causal scene dynamics.
A three-stage training pipeline (visual decoder pretraining, main model warmup, auxiliary decoder warmup, joint fine-tuning) ensures stable optimization.
At inference, auxiliary decoders are discarded and latent tokens are prefilled in a single parallel pass, matching answer-only latency.
Across four benchmarks (NAVSIM, ROADWork, Impromptu, Alpamayo-R1), OneVL is the first latent CoT method to surpass explicit CoT, achieving superior accuracy at answer-only latency.
For example, on NAVSIM it achieves 88.84 PDM-score with 4.46s latency, outperforming AR CoT+Answer (88.29, 6.58s) and prior latent CoT methods.
Ablations show the visual decoder contributes +0.87 PDM-score, language decoder +0.31, and staged training is essential (removing it drops performance to 67.13).
An MLP head variant reaches 86.83 PDM-score at 0.24s latency for real-time deployment.
02 From the paper
Abstract
Chain-of-Thought (CoT) reasoning has become a powerful driver of trajectory prediction in VLA-based autonomous driving, yet its autoregressive nature imposes a latency cost that is prohibitive for real-time deployment. Latent CoT methods attempt to close this gap by compressing reasoning into continuous hidden states, but consistently fall short of their explicit counterparts. We suggest that this is due to purely linguistic latent representations compressing a symbolic abstraction of the world, rather than the causal dynamics that actually govern driving. Thus, we present OneVL (One-step latent reasoning and planning with Vision-Language explanations), a unified VLA and World Model framework that routes reasoning through compact latent tokens supervised by dual auxiliary decoders. Alongside a language decoder that reconstructs text CoT, we introduce a visual world model decoder that predicts future-frame tokens, forcing the latent space to internalize the causal dynamics of road geometry, agent motion, and environmental change. A three-stage training pipeline progressively aligns these latents with trajectory, language, and visual objectives, ensuring stable joint optimization. In inference, the auxiliary decoders are discarded, and all latent tokens are prefilled in a single parallel pass, matching the speed of answer-only prediction. Across four benchmarks, OneVL becomes the first latent CoT method to surpass explicit CoT, delivering superior accuracy at answer-only latency. These results show that with world model supervision, latent CoT produces more generalizable representations than verbose token-by-token reasoning. Code has been open-sourced to the community. Project Page: https://xiaomi-embodied-intelligence.github.io/OneVL