MS Robotics student at Carnegie Mellon's Robotics Institute, working on post-training flow-matching foundation models with Reinforcement Learning in robotic manipulation tasks.
I care about two things: making robots understand the 3D world, and making them act intelligently within it. That usually means working somewhere in the intersection of:
- Reinforcement Learning — policy learning, GRPO, reward shaping for long-horizon tasks
- 3D Perception — Gaussian Splatting, multiview geometry and LiDAR/camera fusion
- Foundation Models for Robotics — VLA fine-tuning, using World Models for trajectory forecasting
Robot sees world in 3D
↓
Builds rich scene representation (3DGS / NeRF / DUST3R)
↓
VLA reasons over scene + language instruction
↓
RL policy executes & improves from interaction
Current focus: improving spatial reasoning in VLAs for dexterous manipulation using learned 3D scene priors.
Occasional notes on things I'm reading or building - RL theory, 3D vision, and robotics systems.
Always happy to talk robots, 3D vision, or RL. Reach out anytime.