Felix Li

University of California, Berkeley

Papers

3

Total Citations

110

H-Index

3

About

Felix Li’s research lies at the critical intersection of reinforcement learning, robotics, and safety, where he tackles two of the field’s most persistent challenges: designing algorithms that can learn from sparse feedback and ensuring reliable behavior under dynamical uncertainty. His most influential contribution is the **SAVED framework** (Safety Augmented Value Estimation from Demonstrations), a model-based RL approach that integrates expert demonstrations to enable safe, sample-efficient policy learning for complex robotic tasks. This work, which has garnered over 90 citations, directly addresses the brittleness of hand-engineered cost functions and the exploration hazards inherent in real-world robotics. Li has also advanced the theory of imitation learning, formalizing the problem of learning from a *converging* supervisor—a scenario where the demonstrator’s policy improves over time, as with a human refining a novel skill. This work, presented at top venues, provides a principled foundation for human-robot collaboration and interactive policy learning. Through his focus on safety-guaranteed, data-efficient algorithms, Felix Li is shaping a future where robots can learn complex behaviors directly from human guidance without compromising on reliability.

Research Focus

Key Achievements

3
H-Index
3
Papers
110
Total Citations
37
Avg Citations/Paper
🏆 Most Cited Paper
Safety Augmented Value Estimation From Demonstrations (SAVED): Safe Deep Model-Based RL for Sparse Cost Robotic Tasks
90 citations · 2020
📈 Most Prolific Year: 2019 (2 Papers)
🤝 Key Collaborators: 10
🏛 Institutions: University of California, Berkeley

Top Papers

  1. 1
  2. 2
  3. 3

Key Collaborators

Contact & Links

Available for collaboration
Content generated · 12 days ago