Xpeng is investing approximately $500 million annually in AI training to enhance its autonomous driving capabilities, claiming to have achieved parity with Tesla's Full Self-Driving (FSD) version 13 and aiming for version 14 soon. The company's VLA 2.0 architecture eliminates language as an intermediate step in driving decisions, focusing on visual data processing to improve efficiency and performance.
Xpeng's strategic decision to eliminate the language translation step in its VLA 2.0 architecture for real-time driving actions is a notable divergence from the industry's trend towards language models. This move aims to reduce unnecessary computation and latency, enhancing efficiency and responsiveness in autonomous driving systems. For those tracking advancements in autonomous vehicle technology, this indicates a potentially more streamlined and effective approach to AI training and model performance, positioning Xpeng as a significant competitor in the race against Tesla's FSD.