Original title: Continuous Actions from Discrete Minds: Latent-Aligned Planning for End-to-End Autonomous Driving
Ruoyu Yao, Yusen Xie, Qingzhao Liu, Pei Liu +4 · September 2026
Self-driving cars must move smoothly in physical space, but modern vision-language AI models think in separate, discrete chunks of information like words. This mismatch makes turning understanding into real-world motion very hard. To fix this, researchers created LaPla, a unified system for self-driving vehicles. First, they built a tool that learns how cars move and maps vehicle paths into a structured space. Then, instead of picking separate chunks that can cause errors, LaPla takes in camera images, past actions, and written text instructions all at once. In a single fast step, it converts these inputs straight into smooth, continuous driving plans. A frozen decoder turns these plans into final physical actions without slow, step-by-step guessing. Testing showed LaPla cut long-horizon planning errors by 15.52% on the nuScenes test and improved driving success rates by 33.34 percentage points in a simulator with much lower delay.
Imagine packing a suitcase where each item must fit perfectly without getting crushed. Normally, you might force items into fixed slots, but some get damaged or don't fit well because the slots are too rigid. Here, the system lets you place items freely in a flexible layout that already knows how to keep everything intact. In this paper, instead of forcing driving actions into fixed codes, it uses a pre-learned space of smooth motions to ensure the car moves naturally.
This matters for passengers and developers of self-driving cars, because it helps the car follow language instructions while moving naturally, reducing errors that could cause accidents. By improving accuracy in long-term planning, it makes autonomous driving more reliable and safer in real conditions.
The abstract doesn't specify limitations in detail, but it only validates the work on the nuScenes benchmark and AlpaSim simulator, which may not cover all driving scenarios or real-world unpredictability.
Ask anything about this paper. Answers come from its abstract, and say so when the abstract doesn't cover your question.
Continuous Actions from Discrete Minds: Latent-Aligned Planning for End-to-End Autonomous Driving — Ruoyu Yao, Yusen Xie, Qingzhao Liu et al.
Abstract licence: CC0
Explanation generated by Synthesis from the abstract — it is not the original paper. This is a preprint, so it hasn't necessarily been peer reviewed.