A research team from Tsinghua University, led by Gao Yang, has developed a new embodied AI model called FP3. This model utilizes pre-training and 3D point cloud representation to achieve over 90% success rate on various robotic tasks, significantly outperforming existing baseline methods. FP3 demonstrates robust performance in zero-shot testing on unseen objects and environments, with an average success rate exceeding 80%. The researchers also highlighted potential improvements, including integrating with large visual-language models and enhancing its 2D feature fusion capabilities. AI
IMPACT Sets a new benchmark for embodied AI task success, potentially accelerating progress in robotics and real-world AI applications.
RANK_REASON The cluster details a new embodied AI model (FP3) and its performance on benchmarks, along with related company advancements and funding. [lever_c_demoted from research: ic=1 ai=1.0]
- Era0
- FP3
- GigaBrain-0.1
- Nvidia
- Physical Intelligence Pi0.5
- Qianxun Intelligent
- RoboArena
- RoboChallenge
- Spirit v1.5
- Spirit v1.6
- Tsinghua University
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
