Research Intern, Scene Understanding and Planning

DiDi • San Jose, CA

4d • Onsite

About The Position

DiDi Autonomous Driving, established in 2016 and made independent as Voyager in 2019, develops Level 4 autonomous driving technology intended to make transportation safer, more efficient, and accessible. The team combines DiDi’s data, operational expertise, and mobility ecosystem to develop and deploy autonomous mobility solutions.

The Research Intern will work on scene understanding and planning for autonomous vehicles, helping the planner maintain an accurate, current understanding of real-world driving environments rather than relying only on static maps. The role involves tackling open research problems from real-world driving, from problem definition through validation. Strong results may be published, with support for writing them up, while the primary objective is to solve problems relevant to the product. The intern reports directly to the team lead and collaborates with production engineers and the Perception, Prediction, Planning, and Mapping teams.

This is a full-time, onsite internship lasting 3–6 months, with flexible start dates. The hourly rate in the selected city is $46.

Requirements

• Be currently enrolled in an M.S. or Ph.D. program in Computer Science, Robotics, Electrical Engineering, Machine Learning, or a related field for the full duration of the internship.
• Be available for a full-time, onsite internship lasting 3–6 months; start dates are flexible.
• Have strong Python programming skills and experience with modern deep learning frameworks such as PyTorch or JAX.
• Have hands-on experience designing, training, and debugging deep learning models for computer vision, 3D perception, or sequential decision-making.
• Have a solid understanding of autonomous driving fundamentals, including perception, prediction, and motion planning.
• Be able to critically read research, form hypotheses, and turn them into well-controlled experiments.
• Have excellent communication and collaboration skills.

Nice To Haves

• Publications at top venues such as CVPR, ICCV, ECCV, NeurIPS, ICLR, ICML, ICRA, IROS, CoRL, or IV.
• Experience with multimodal sensor fusion (camera, lidar, radar) or bird’s-eye-view (BEV) representations.
• Experience with transformers, graph neural networks, or other structured scene representations.
• Experience with uncertainty estimation or learning-based prediction or planning.
• Experience with large-scale model training and data pipelines.
• Experience with public autonomous driving datasets such as nuScenes, Argoverse 2, or the Waymo Open Dataset.
• Proficiency in C++; CUDA experience is a plus.
• Experience with robotics systems.

Responsibilities

• Survey current research in scene understanding, prediction, and planning, and develop practical approaches suited to the system and its constraints.
• Design and implement deep learning models that build real-time understanding of the driving environment using onboard sensors and other available sources.
• Train and evaluate models using large-scale driving-log data and public benchmarks.
• Lead failure analysis to determine where and why models fall short.
• Define metrics and evaluation methods that connect perception quality to vehicle driving performance.
• Integrate successful prototypes with the planning stack, with support from production engineers.
• Collaborate with Perception, Prediction, Planning, and Mapping teams to move research results from experiments into products.

What This Job Offers

• Job Type: Full-time
• Career Level: Intern