Trustworthy Robotics
ChatGPT 4oThe brain of AI has been replaced by large language models (LLMs). The next challenge (or long-standing challenge) is how to build Physical AI? We attach this challenge by making Physical AI trustworthy. In particular, we consider the following question.
How to build safe robots or humanoids that could live with us and even our kids?
Based on our previous experiments on trustworthy AI, we build ground breaking learning methods to achieve this goal.
On-going/Potential Projects
- Safe Humanoid Learning: Learn safe humanoids.
- Policy Calibration: Calibrate RL policies.
- Safe RL with Certification: Guarantee safety in RL.
- Adversarially Robust Policy: Learn a secure policy from jailbreaking.
What’s your own project idea?
Keywords
- Vision-Language-Action Models (VLAs)
- Reinforcement Learning (RL)
- Selective Prediction
- Conformal Prediction
- Uncertainty Quantification