Reproduce “AutoVLA-D: Learn a Trajectory-Geometric Condition from Reasoning for Autonomous Driving VLA” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Yuguang Yang, Zhewen Tan, Canyu Chen, Linlin Yang, Cheng Chi, Chunyang Liu, Kehua Sheng, Jinyu Yang and 4 more
AutoVLA-D uses discrete motion-shaped action tokens as an intermediate condition for generating continuous driving trajectories, rather than as final outputs. In a 768-token NAVSIM navtest ablation with Qwen3-VL-4B, Brohan TrajCond scored 85.69 EPDMS, above 83.19 for continuous conditioning; direct Brohan output scored 83.00 versus 83.90 for continuous output. These benchmark comparisons do not establish gains in every driving setting.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.