Ant Group's AI lab has open-sourced LingBot-Vision, a foundational vision model designed for embodied AI, and LingBot-Depth 2.0, a spatial perception model built upon it. These models aim to significantly improve how robots perceive and understand their physical environment, particularly in challenging scenarios involving transparent or reflective objects, small or distant targets, and complex indoor settings. LingBot-Vision's unique approach focuses on learning object boundaries and spatial structures during pre-training, enabling it to achieve high accuracy with a smaller parameter count compared to other large vision models. AI
IMPACT Enhances robot capabilities in real-world tasks by improving spatial understanding and perception accuracy.
RANK_REASON Open-source release of foundational models for embodied AI. [lever_c_demoted from research: ic=1 ai=1.0]
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
