ML Engineer, Motion Planning

DiDi Labs•San Jose, CA

5d•$169,783 - $338,694

About The Position

DiDi Autonomous Driving, which became the independent company Voyager in 2019, develops Level 4 autonomous driving technology. The ML Engineer will join the autonomous vehicle team and develop the next-generation machine-learning-powered planning system for Voyager’s autonomous driving platform. The role uses frontier machine learning methods, including learned planners and end-to-end models, to understand complex driving environments, make data-driven decisions, and navigate safely and efficiently. It also leverages real-world driving data to improve model performance, create evaluation frameworks related to on-road safety, and support deployment and scaling of driverless technology. The position combines research and production engineering: framing real-world driving challenges as machine-learning problems, building and evaluating models with driving data and simulation, integrating learned methods with traditional planning, and deploying systems within onboard compute, latency, and safety constraints. The engineer will work with teams across infrastructure, perception, prediction, planning, mapping, and systems engineering.

Requirements

B.S., M.S., or Ph.D. in Computer Science, Machine Learning, Robotics, Electrical Engineering, Mechanical Engineering, or a related technical field, or equivalent practical experience.
At least 3 years of industry experience for Senior level or 5 years for Staff level in machine learning, autonomous driving, robotics, or related fields.
Proven experience developing and deploying ML models in production, ideally for real-time, embedded, or autonomous driving systems.
Strong proficiency in Python and C++, production software development experience, and hands-on experience with modern deep-learning frameworks such as PyTorch, JAX, or TensorFlow.
Deep understanding of modern deep-learning techniques, including transformers, BEV and 3D representations, and trajectory prediction or generation models, with hands-on model training, evaluation, and debugging experience.
Foundational knowledge of linear algebra, control theory, or motion planning, including path and trajectory generation, trajectory evaluation, cost-function design, and lane and road geometry.
Experience applying machine learning to motion planning, robotics, or related decision-making problems and integrating learning-based approaches with traditional planning methods.
Ability to own ambiguous technical challenges end to end while balancing research innovation with production constraints, safety, reliability, and engineering trade-offs.
Demonstrated technical impact through research publications, open-source contributions, or significant industry achievements.

Nice To Haves

Strong engineering experience applying ML-driven approaches to complex real-world problems and translating research ideas into robust, production-quality systems.
Experience integrating learning-based approaches with traditional rule-based or optimization-based planning systems to improve safety, efficiency, and robustness.
Experience with closed-loop simulation, large-scale evaluation, and data-driven model iteration for autonomous driving.
Experience mentoring junior engineers, supervising research interns, or leading small technical teams.
Experience with road-structure representations, including lane graphs, topology reasoning, scene understanding, or uncertainty estimation in complex driving environments.
Experience with end-to-end autonomous driving models, learned planners, vision-language models, or foundation models applied to driving and robotics.
Experience deploying and optimizing ML models for real-time onboard inference, including TensorRT, CUDA, quantization, and compute or latency optimization.
Publications at leading conferences such as CVPR, ICCV, ECCV, NeurIPS, CoRL, ICRA, IROS, or IV, or equivalent demonstrated technical impact in industry.

Responsibilities

Develop ML-powered systems for driving behavior decisions, scene understanding, and trajectory generation to support Level 4 autonomous driving deployment.
Frame real-world driving challenges as machine-learning problems and research and apply techniques such as imitation learning, reinforcement learning, learned planners, and end-to-end models.
Design data-driven algorithms to improve autonomous driving safety, generalization, and reliability, including scene understanding and planning in complex or changing road environments.
Develop hybrid planning solutions that combine machine-learning methods with traditional rule-based and optimization-based planning.
Develop and evaluate large-scale ML models using real-world driving data and simulation, and integrate models into production systems subject to onboard compute, latency, and safety constraints.
Build rapid-iteration infrastructure, including challenging-scenario mining, training and evaluation pipelines, closed-loop simulation, and metrics correlated with real-world driving performance.
Collaborate with ML Infrastructure, Perception, Prediction, Planning, Mapping, and Systems Engineering teams to investigate failures, drive end-to-end improvements, and deploy solutions.