Robotic Reinforcement Learning · AI Agents · LLM Infrastructure · Sim-to-Real

PhD Student — Open to Research / Robotics / ML Internships.

PhD student working at the intersection of Robot Learning and Large-scale Systems — I train RL policies in simulation and push them toward real hardware, and I care about the infrastructure that makes both fast and correct.

I came to AI from an unusual direction: a B.Arch / architecture background that taught me to reason about complex systems, geometry, spatial relationships, and the tradeoffs between elegant ideas and real-world constraints. Over time, I became convinced that many of the most important problems in design and the physical world will be solved not by better static tools, but by intelligent systems that can learn, adapt, and act. That belief drew me to robot learning, simulation, and AI infrastructure. I am especially energized by unfamiliar, technically demanding problems and by the process of turning ideas across disciplines into scalable systems that work in practice.

• 🤖 Robotics RL in Simulation — building and training policies in NVIDIA Isaac Sim / Isaac Lab (locomotion / manipulation / sim-to-real) with GPU-parallel environments and PPO/SAC-style training.

• ⚙️ LLM infrastructure — contributor to vLLM, the core LLM inference engine: 2 performance PRs merged.

• 🦾 ROS 2 — bridging learned policies to real robot stacks (nodes, controllers, perception → action pipelines).

• 🔬 Research interests: Reinforcement Learning, AI Agent, Robot Manipulation/Locomotion, GPU-accelerated simulation and sim-to-real transfer.

📫 lynnhe02@gmail.com · 📍 Texas

• PR #46542 — [Perf][LoRA] (merged): Replaced a per-token list.index() lookup in convert_mapping — an O(num_tokens × num_loras) hot path the code had flagged with a TODO — by building a reverse {lora_id: index} dict once for O(1) lookups, cutting mapping construction to O(num_tokens). 2.5×–6.5× faster in microbenchmarks (e.g. 64 LoRAs / 1024 tokens: 275µs → 42µs), with identical output verified against randomized + existing LoRA tests.
• PR #46543 — [Perf][Multimodal] (merged): Removed a wasteful O(num_frames) timestamp-list allocation in GLM-4V / GLM video frame sampling, computing each timestamp inline as frame_index * duration_per_frame. Byte-for-byte identical behavior with lower memory on long videos.

Robotics & Simulation

AI / ML

Languages & Systems