In a remarkable breakthrough for autonomous driving technology, XPENG and Peking University have been recognized at the prestigious AAAI 2026 conference with a pioneering research paper that could transform how artificial intelligence processes visual information in self-driving vehicles.

The collaborative research, titled "FastDriveVLA: Efficient End-to-End Driving via Plug-and-Play Reconstruction-based Token Pruning", was selected from an incredibly competitive pool of 23,680 submissions, with only 4,167 papers accepted – representing a highly selective 17.6% acceptance rate. This achievement underscores the cutting-edge nature of XPENG's approach to autonomous driving artificial intelligence.

At the heart of their innovation is FastDriveVLA, a sophisticated visual token pruning framework designed to mimic human visual perception. The method enables AI systems to focus exclusively on critical visual information, effectively filtering out irrelevant background data – much like how human drivers instinctively concentrate on lanes, vehicles, and pedestrians while navigating complex traffic scenarios.

As Vision-Language-Action (VLA) models become increasingly sophisticated, processing visual tokens has emerged as a significant computational challenge. Traditional approaches struggled with efficiently managing the massive amount of visual data generated during autonomous driving. XPENG and Peking University's breakthrough introduces an adversarial foreground-background reconstruction strategy that dramatically reduces computational load without compromising decision-making accuracy.

In practical tests on the nuScenes autonomous driving benchmark, the framework demonstrated remarkable capabilities. By reducing visual tokens from 3,249 to just 812, FastDriveVLA achieved a stunning 7.5x reduction in computational complexity while maintaining high planning precision. This represents a significant leap forward in making autonomous driving systems more efficient and responsive.

This isn't XPENG's first major AI recognition in 2026. Previously, they were the sole Chinese automaker invited to speak at CVPR WAD, showcasing their commitment to pushing technological boundaries. Their November Tech Day further highlighted their innovative spirit by unveiling a VLA 2.0 architecture that eliminates traditional language translation steps, enabling direct Visual-to-Action generation.

XPENG's long-term vision extends beyond individual technological achievements. As an 'Explorer of Future Mobility', the company is steadfastly committed to achieving Level 4 autonomous driving. With R&D centers spanning Guangzhou, Beijing, Shanghai, and international locations like the United States, XPENG continues to invest heavily in AI large model technology, aiming to seamlessly integrate intelligent systems into vehicles.

By prioritizing full-stack in-house development and maintaining a global research perspective, XPENG is not just developing technology – they're reshaping the future of transportation, one innovative breakthrough at a time.