XPENG–Peking University Collaborative Research Accepted By AAAI 2026: Introducing A Novel Visual Token Pruning Framework For Autonomous Driving - CleanTechnica

XPENG, in collaboration with Peking University, has developed FastDriveVLA, a novel visual token pruning framework designed to enhance autonomous driving AI by enabling it to focus on essential visual information, mimicking human driving attention. This approach significantly reduces computational load—by approximately 7.5 times—while maintaining high planning accuracy. The framework employs an adversarial foreground-background reconstruction strategy to effectively identify and retain critical tokens related to lanes, vehicles, and pedestrians, filtering out irrelevant background data. FastDriveVLA demonstrated state-of-the-art performance on the nuScenes autonomous driving benchmark, reducing visual tokens from 3,249 to 812 without compromising driving decisions. The research paper detailing FastDriveVLA was accepted by AAAI 2026, a leading artificial intelligence conference with a competitive acceptance rate of 17.6%. This recognition highlights XPENG’s advanced capabilities in AI-driven mobility and their commitment to accelerating Level 4 autonomous driving. XPENG’s recent achievements include presenting at CVPR WAD