This summer, I had the opportunity to participate in the DOE SULI program under my mentor Wei Xu at Brookhaven National Laboratory, where I worked on training the humanoid robots using reinforcement learning in simulation.
🤖 My project focused on Unitree G1 and H1 robots, developing a modular Markov Decision Process (MDP) training framework in NVIDIA Isaac Lab, powered by the skrl RL library and the Proximal Policy Optimization (PPO) algorithm.
🔧 Key components of my work included:
• 🧩 Extending Isaac Lab’s modular MDP framework for specific locomotion/manipulation tasks
• 📦 Implementing reinforcement learning pipelines with skrl
• 📈 Tuning PPO hyperparameters for stable and efficient locomotion training
• 🦿 Simulating control behaviors for humanoid and robots
• 🔁 Structuring code for reusability in future robotics projects
This framework enables rapid experimentation with different robot types, control policies, and task environments, supporting advancements in robotics research, simulation-to-real transfer, and intelligent control systems.
🧠 Skills gained: Python, NVIDIA Isaac Lab, NVIDIA Isaac Sim, reinforcement learning, PPO, robotics simulation
I’m so grateful to my mentor, Dr. Xu, the WDSE department, DOE, and BNL for an amazing experience, and I’m excited to bring these skills into my future work in robotics, AI, and computational science!
#Robotics #ReinforcementLearning #IsaacLab #skrl #PPO #Unitree #DOE #SULI #BrookhavenNationalLab #AI #Simulation #HumanoidRobotics #Unitree #Internship #Research #ComputationalScience
