cleantechnica.com
Nov 6, 2026
XPENG Researchers and Academic Partners Introduce GroundingPI, An Open-Source Visual Perception Model For Physical AI
AI-summarised brief · reviewed before publication
XPENG, in partnership with Peking University, the University of Hong Kong, and other academic institutions, released GroundingPI, a 4‑billion‑parameter open‑source visual perception model. The model excels in 34 benchmarks across 11 perception categories, achieving the highest aggregate score of 73.68 among 44 baseline models. GroundingPI focuses on precise object identification and spatial localization, providing point and bounding‑box coordinates for real‑world tasks. Its code, weights, and research paper are publicly available, enabling further development of embodied AI systems.
💡 Why It Matters
- · By outperforming larger general‑purpose models on specialized visual grounding tasks, GroundingPI demonstrates that targeted perception capabilities can accelerate the deployment of reliable robotics and autonomous vehicles.
