LLM trained locomotion policy deployed zero-shot on challenging environments
Humanoid Locomotion using Reinforcement Learning and AI Agents
RoMeLa
I pioneered a fully automated reinforcement learning framework for humanoid locomotion that combines GPU-accelerated MJX simulation, accurate modeling of parallel mechanisms, and multi-agent LLMs using retrieval-augmented generation (RAG) for curriculum design. This system dynamically generates and refines training stages based on performance, enabling robust policy learning without manual tuning. The resulting policies achieved zero-shot transfer to hardware, where the BRUCE robot demonstrated stable walking outdoors and successfully handled complex transitions like stepping off elevated platforms, highlighting the framework’s real-world generalization and autonomy. My future works include integration imitation learning to the framework.
Advanced Robotic Perception Systems for Humanoid Soccer
RoMeLa
I developed the full perception stack for the humanoid robot ARTEMIS, enabling full spatial awareness and advanced decision-making in dynamic RoboCup soccer environments. This stack integrated the YOLOv8 deep learning model and classical computer vision algorithms with point clouds for object detection, 3D pose estimation, and trajectory prediction robust to heavy amounts of noise.
Object Segmentation using Vision Transformers and Deep Learning models
RoMeLa
I integrated the Segment Anything Model (SAM) vision transformer with custom YOLOv8 detection weights to provide 95% accurate segmentation of slide handles and stairs. The segmented object's positions are extracted from the Intel Realsense D435 camera's point cloud for use in simultaneous locomotion and grasping.
Cost Efficient 3D Printed Robot Dog
Evodyne Robotics
The robot dog project involved designing and developing a fully functional quadruped robot. I 3D modeled and manufactured the upgraded big dog, ensuring a compact and efficient mechanical design optimized for strength and cost efficiency. I implemented a PID control system integrated with an IMU to enable real-time balance and stability. This combination of mechanical and control systems engineering allowed the robot to trot and balance on a tilting platform while being powered by cheap servos to lower production cost.
First Tech Challenge Robotics
9614 Hyperion Team Captain
I captained my FTC robotics team for two years, leading the design, assembly, and programming of competition-ready robots. I spearheaded the development of both teleoperated and autonomous functionalities, utilizing advanced techniques such as Bezier Curve Pure Pursuit for precise path planning and computer vision for object detection and navigation. We competed in the championship rounds for regional competitions and placed in the top 20 internationally, showcasing our team's innovative designs and technical expertise.