https://www.morningstar.com/news/pr-newswire/20251229cn54116/xpengpeking-university-collaborative-research-accepted-by-aaai--introducing-a-novel-visual-token-pruning-framework-for-autonomous-driving
XPENG, in collaboration with Peking University, has developed FastDriveVLA—a novel visual token pruning framework that enables autonomous driving AI to “drive like a human” by focusing only on essential information, achieving a 7.5x reduction in computational load.
The research has been accepted by AAAI 2026, one of the world’s premier AI conferences, which had a highly selective acceptance rate of just 17.6% this year.
This achievement underscores XPENG’s full-stack capabilities in AI-driven mobility and advances the industry toward efficient, scalable deployment of next-generation autonomous driving systems.
FastDriveVLA Reduces Computational Load by 7.5x While Retaining Accuracy
Most existing AI driving frameworks are bogged down by massive computational demands due to the sheer number of visual tokens processed by autonomous vehicle models. FastDriveVLA changes the game by pruning away unnecessary tokens using a human-inspired approach: focusing only on truly relevant visual cues.
AAAI’s rigorous acceptance criteria highlight just how novel FastDriveVLA is among global AI research. Out of 23,680 submissions to AAAI 2026, only 4,167 papers were accepted, a mere 17.6% of entries. XPENG’s work not only broke into this elite group but did so with a solution that directly tackles real-world bottlenecks in deploying L4-level driverless systems.
The 6 Levels of Vehicle Autonomy Explained
Level 4 (High Driving Automation)
The key difference between Level 3 and Level 4 automation is that Level 4 vehicles can intervene if things go wrong or there is a system failure. In this sense, these cars do not require human interaction in most circumstances. However, a human still has the option to manually override.
But until legislation and infrastructure evolves, they can only do so within a limited area (usually an urban environment where top speeds reach an average of 30mph). This is known as geofencing. As such, most Level 4 vehicles in existence are geared toward ridesharing.
Level 5 (Full Driving Automation)
Level 5 vehicles do not require human attention―the “dynamic driving task” is eliminated. Level 5 cars won’t even have steering wheels or acceleration/braking pedals. They will be free from geofencing, able to go anywhere and do anything that an experienced human driver can do.
Top 3 U.S. Level 4 Self-Driving Cars You Can Buy in 2026
- Tensor Robocar
- Lucid Gravity
- Tesla Cybercab
Xpeng’s L4-capable vehicles are set to enter mass production in 2026, with pilot Robotaxi operations and services launching in select regions. Xpeng’s humanoid robot is expected to enter mass production in the second half of 2026.
Xpeng plans to start producing vehicles with L4 capabilities and to run robotaxi tests in selected areas next year. After securing regulatory approval, the company expects to conduct trials independently and then engage potential partners to explore commercial deployment
I believe Tesla was hoping all EV manufacturers would adopt its FSD.
According to Wikipedia[1] there are 40 million EVs in use on the earth. 21.8 million in China. I expect China will go with a domestic Autonomous Driving system rather than a foreign entity. So that shaves 50% of potential autonomous driving market. And I expect China companies will aggressively promote their version of FSD; just as they have promoted aggressively their EV models.
And Xpeng plans to introduce their humanoid robot next year too!