Robotics researcher in UCLA's Robotics and Mechanisms Laboratory, working on DexDirect — a more efficient, scalable approach to collecting demonstrations and training policies for dexterous robotic hands.
- Led kinesthetic design and prototyping for direct-guidance demonstration collection on dexterous hands (e.g. MIDAS)
- Recorded task demonstrations across image, state, and tactile observations and trained a neural network policy via imitation learning
- Monitored policy rollout on hardware and tuned executor-side firmware to leverage prediction consistency
- DexDirect website link: https://dexdirect.github.io/dexdirect/
B.J. Kim*, S.J. Wang, et al. — DexDirect: Direct Kinesthetic Arm Guidance for Efficient Dexterous Demonstration Collection. Approved for 2026 IEEE-RAS 25th International Conference on Humanoid Robots on Oct. 7, 2026.