YLX中

Autonomous Driving · Humanoid Robotics · Legged Systems

Multimodal Reinforcement Learning

Reinforcement Learning & Embodied Intelligence

← Back

Selected Projects

🚗 Autonomous Driving with Reinforcement Learning

Keywords: PPO · CARLA · Multi-Agent Learning · Distributed Training

This project focuses on developing a reinforcement learning-based autonomous driving framework in high-fidelity simulation environments.

  • Built on the CARLA simulator to model realistic urban driving scenarios
  • Designed a multi-modal observation space combining visual input, vehicle state, and route information
  • Implemented Proximal Policy Optimization (PPO) for policy learning
  • Introduced multi-agent adversarial training to improve robustness in dynamic traffic environments
  • Integrated distributed sampling and training using Ray RLlib to accelerate convergence

Highlights

  • Improved training efficiency through parallelized experience collection
  • Enhanced policy robustness under complex multi-vehicle interactions
  • Designed physically-informed reward functions for stability and controllability

🤖 Humanoid Robot Locomotion with Tactile Feedback

Keywords: Reinforcement Learning · Tactile Sensing · Gait Optimization · Sim-to-Real

This project explores reinforcement learning for humanoid locomotion with explicit modeling of physical contact feedback.

  • Developed a learning framework incorporating foot pressure distribution from tactile sensors (e-skin)
  • Designed reward functions based on contact stability and force distribution
  • Modeled the relationship between physical interaction and control policy optimization
  • Investigated sim-to-real transfer by embedding physically meaningful feedback signals

Highlights

  • Improved locomotion stability through contact-aware learning
  • Bridged the gap between simulation and real-world dynamics
  • Emphasized physically grounded policy learning rather than purely data-driven strategies

🦿 Wheel-Legged Robot Control (Course Project)

Keywords: Hybrid Locomotion · Dynamics Modeling · Control System Design

This course project focuses on the modeling and control of a wheel-legged robotic system.

  • Built a dynamic model of a wheel-legged robot considering coupled motion constraints
  • Designed control strategies for stable locomotion under hybrid motion modes
  • Implemented trajectory tracking and posture stabilization
  • Analyzed system behavior under different motion conditions

Highlights

  • Developed understanding of hybrid locomotion systems
  • Combined classical control with system modeling
  • Strengthened intuition on physical constraints in robotic systems

💡 Research Direction

My current research interest lies in:

  • Reinforcement learning for control optimization
  • Physically grounded learning systems
  • Sim-to-real transfer in embodied intelligence
  • Multi-agent interaction and adaptive optimization

I aim to develop intelligent systems that integrate perception, decision-making, and physical consistency, enabling robust performance in real-world environments.