Chris Zhang

prof_pic_new.jpg

I am currently a 5th year PhD student in the Machine Learning Group at the University of Toronto, advised by Raquel Urtasun. Prior to this, I obtained my BASc in Systems Design Engineering at the University of Waterloo.

I am also currently a Research Scientist at the self-driving startup Waabi. Here I research learning-based driving policies, with applications in behavior simulation and autonomy.

My goal is to develop safe and intelligent agents for the real world. Key research questions I’m interested in are:

  • What is the right data engine to enable scalable learning?
  • What learning algorithms enable effective scaling with compute?
  • How can we guarantee, test or validate the safety and performance agents we deploy?

publications

  1. scenecontrol.png
    SceneControl: Diffusion for Controllable Traffic Scene Generation
    Jack Lu*, Kelvin Wong* , Chris Zhang, Simon Suo, and Raquel Urtasun
    In International Conference on Robotics and Automation (ICRA) , 2024

    Use guided sampling to generate realistic, constraint-satisfying traffic initialization scenes.

  2. rtr1.png
    Learning Realistic Traffic Agents in Closed-loop
    Chris Zhang, James Tu, Lunjun Zhang, Kelvin Wong, Simon Suo, and Raquel Urtasun
    In 7th Annual Conference on Robot Learning (CoRL) , 2023
    arXiv Website

    Make imitation-learning driving policies more robust with reinforcement learning and synthetic long-tail scenarios.

  3. guard1.png
    Towards Scalable Coverage-Based Testing of Autonomous Vehicles
    James Tu, Simon Suo , Chris Zhang, Kelvin Wong, and Raquel Urtasun
    In 7th Annual Conference on Robot Learning (CoRL) , 2023
    Website

    Formulate structured scenario-based testing as a level-set estimation problem and use Gaussian Processes to obtain probabilistic coverage estimates.

  4. travl.png
    Rethinking Closed-loop Training for Autonomous Driving
    Chris Zhang*, Runsheng Guo*, Wenyuan Zeng*, Yuwen Xiong, Binbin Dai, Rui Hu, Mengye Ren, and Raquel Urtasun
    In European Conference on Computer Vision (ECCV) , 2022
    arXiv Website

    Learn explicit long-horizon planners with data-efficient reinforcement learning and study the effects of the training scenarios on learned policies.

  5. ghn.png
    Graph HyperNetworks for Neural Architecture Search
    Chris Zhang, Mengye Ren, and Raquel Urtasun
    In International Conference on Learning Representations (ICLR) , 2019
    arXiv

    Speed up architecture search by learning to predict the weights of a candidate architecture by message passing along its computation graph.

  6. 3dv2.png
    Efficient Convolutions for Real-time Semantic Segmentation of 3d Point Clouds
    Chris Zhang, Wenjie Luo, and Raquel Urtasun
    In 2018 International Conference on 3D Vision (3DV) , 2018
    PDF

    Replace 3D convolutions with 2D convolutions in birds eye view for more efficient semantic segmentation.