Research Projects
Here are some of my past research projects in reinforcement learning, robotics, and AI applications.
Fault-Tolerant Locomotion for Quadruped Robots
In this project, I focused on developing fault-tolerant locomotion controllers for quadruped robots (Unitree Go2). We proposed a latent-state reinforcement learning framework that removes the need for explicit fault detection. To improve robustness, I introduced a curriculum-based training strategy that gradually increases the number and types of motor faults during training, enabling stable locomotion under weakened, locked, and zero-torque joint conditions. The project is still ongoing.
Watch the demos:
Vision-Guided Navigation–Locomotion Integration
In this project, I contributed to integrating vision-based terrain perception with locomotion control to improve adaptability in complex environments. My work involved incorporating elevation maps and traversability predictions into the locomotion pipeline, enabling the policy to anticipate terrain challenges and adjust gait patterns accordingly. This contribution supported the broader goal of combining navigation and fault-tolerant locomotion, bridging perception and control for more robust real-world deployment.
Watch the demo:
Humanoid Robot Motion Imitation
Developed a control strategy for the humanoid robot G1 using imitation learning, where a well-designed reward function enables the robot to accurately replicate the dynamic motion capture data of professional dancers.
Watch the demo:
Bipedal Robot Locomotion using Reinforcement Learning
Implemented a PPO-based stable walking algorithm for the bipedal robot Hector in Isaac Gym, utilizing an asymmetric actor-critic approach.
Watch the demo:
Resource Allocation and Scheduling in IoT & Edge Computing
Designed a deep reinforcement learning algorithm for efficient microservice scheduling in edge computing environments, balancing latency and resource constraints.
System Architecture: 
Time Slot Design: 
YOLO-based Smoking Detection
Personal Project(for fun)
Developed a real-time smoking behavior detection system using YOLO object detection model. The system can accurately detect and classify smoking activities in various environments.
Watch the detection demo: