Wenchao Ding

I worked on a comprehensive product-level prediction solution for autonomous vehicles in complex urban environments. I also led a pioneer team working on the problem of how to achieve the best combination of prediction and decision. We proposed "Coupled Prediction & Decision Networks". - Research on coupled prediction & decision networks: I led a pioneer team working on several key problems for urban autonomous driving: 1) how to combine prediction and planning; 2) how to design the planning module in a scalable (learnable) way; 3) how to achieve human-like decision-making; 4) how to integrate neural nets with traditional model-based planners. Our methods are validated on-board in the toughest real-world scenarios and illustrate promising performance. - Comprehensive product-level prediction solution: I designed a comprehensive learning-based prediction solution for trajectory and intention prediction in urban environments. For mass production, I also designed a scalable data pipeline (data mining, circling, and finetuning) for constantly improving the prediction neural nets.

My second internship at DJI Automotive. I deployed my latest decision-making algorithms onboard and tested them in dense real-world traffic in one of the most cluttered cities in China. - I worked on novel decision-making (model-based) algorithms for autonomous vehicles in real-world dense city traffic. Our algorithms can achieve safe, comfortable, and socially compliant merging and lane changing behaviors in dense traffic. - See our EPSILON paper "EPSILON: An Efficient Planning System for Automated Vehicles in Highly Interactive Environments" in IEEE Transactions on Robotics (TRO) and EUDM paper "Efficient Uncertainty-aware Decision-making for Autonomous Vehicles Using Guided Branching" in ICRA 2020.

I worked in the HKUST Aerial Robotics Group under the supervision of Prof. Shaojie Shen. I am passionate about studying planning problems that have real-world impact.

During my PhD, I spent much of my time in DJI Automotive to deploy my latest research onboard and try my algorithms in the real world. - I propose a motion planning algorithm named spatio-temporal corridor for autonomous vehicles, which is a unified motion planning solution for complex urban environments. - See our RAL paper "Safe Trajectory Generation For Complex Urban Environments Using Spatio-temporal Semantic Corridor" for technical details.