Sumit Kumar

Sumit Kumar

Staff Tech Lead Manager at Aurora Innovation

About Me

I am a Staff Tech Lead Manager at Aurora Innovation, where I lead the Intent Prediction team within the Behavior Planning organization. My team owns the core ML forecasting component that powers Aurora’s autonomous vehicles, including architecture decisions, roadmap, and cross-functional alignment across Motion Planning, Perception, Maps, Simulation, and Testing.

Previously, I worked as a Software Engineer at Aurora Innovation (and UATC LLC, which was acquired by Aurora), where I developed GNN-based scene reasoning models for motion planning. I hold a Master of Science in Robotics from Carnegie Mellon University (advised by Prof. Katia Sycara) and a B.Tech in Mechanical Engineering with a Minor in Artificial Intelligence from IIT Kanpur. Check out my Resume and Google Scholar for more details.

Patents

  • Systems and Methods for Interaction-Based Trajectory Prediction — US Patent No. 11,975,726 (Issued May 2024)
  • Goal-Based Motion Forecasting — US Patent No. 11,801,871 (Issued Oct 2023)
  • Autonomous Vehicle Motion Planning — US Patent Application No. 63/616,284 (Under Review)
  • Training Motion Planning with Hybrid Simulation — US Patent Application No. 19/433,843 (Under Review)

Publications

  • Interaction-Based Trajectory Prediction Over a Hybrid Traffic Graph [Paper]
    Sumit Kumar, Yiming Gu, Jerrick Hoang, Galen Clark Haynes, Micol Marchetti-Bowick
    IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS 2021)

  • MAME: Model Agnostic Meta Exploration [Paper] [Code]
    Swaminathan Gurumurthy, Sumit Kumar, Katia Sycara
    Conference on Robot Learning (CoRL 2019)
    This paper was selected for an oral presentation at the Adaptive and Multitask Learning: Algorithms & Systems Workshop at ICML 2019.

  • Learning Transferable Cooperative Behavior in Multi-Agent Teams [Paper] [Code]
    Akshat Agarwal*, Sumit Kumar*, Katia Sycara
    This paper was selected for a spotlight presentation at the Learning and Reasoning with Graph-Structured Data Workshop at ICML 2019.

  • Active Learning with Gaussian Processes for High Throughput Phenotyping [Paper] [Code]
    Sumit Kumar, Wenhao Luo, George Kantor, Katia Sycara
    International Conference on Autonomous Agents and Multiagent Systems (AAMAS 2019)