Zhouheng Li

During my internship as an End-to-End Autonomous Driving Algorithm Intern at Robonets, I worked on safe multi-agent motion planning for autonomous transportation in mining scenarios. To address the challenges of limited space in loading areas and complex vehicle interactions,I designed and implemented a decentralized trajectory generation module based on a lightweight diffusion model, incorporating road topology, target poses, and predicted trajectories of nearby vehicles as planning conditions. I also developed a trajectory refinement method using model predictive projection control to smooth the generated trajectories and enforce kinematic constraints. In addition, I built a simulation-based validation workflow and analyzed metrics such as traffic efficiency, collision risk, minimum distance, tracking error, and inference latency to evaluate the proposed framework’s safety, real-time performance, and trajectory quality.

During my internship as a Motion Planning Algorithm Engineer for robotic manipulators, I was mainly involved in optimization-based motion planning tasks. Working within the ROS framework, I used Python and C++ to organize, implement, and debug functional modules, while supporting system integration and validation. I also implemented trajectory smoothing for robotic manipulators using quadratic programming in Python, and collected IMU data to explore trajectory reconstruction methods. In addition, I summarized commonly used metrics for evaluating manipulator trajectory performance, including trajectory smoothness, execution time, path length, joint velocity, and acceleration variations, providing useful references for subsequent trajectory analysis and algorithm comparison.

Focusing on VLA-enhanced multi-agent game planning for autonomous racing