Autonomous driving research — XCoT-VLA

Organisation/country: XPeng, China. System: vision-language-action driving model. AI technology: compact executable reasoning tokens conditioned jointly with trajectory queries. Autonomy: research-stage planning component, not proof of road-ready autonomous driving. Purpose: reduce verbose reasoning overhead while linking perception and trajectory generation. Authors report improved trajectory errors and real-time budgets; the work is an unreproduced preprint. arXiv