XPeng backed GroundingPI, an open-source visual perception foundation model. Researchers affiliated with XPeng and academic institutions developed the model. GroundingPI enables AI systems to identify objects and understand their spatial positions. This capability is key for robotics and autonomous driving.
The model features 4 billion parameters. It reportedly outperformed GPT-6 Astra. The research team's evaluation covered 34 perception benchmarks.
XPeng and its partners made the research paper, model weights, and code publicly available to foster further research and development in perception-focused Physical AI. This could enhance future autonomous systems.