Senior ML Engineer (Motion Planning)

Do you want your code to control real robots tomorrow? Our delivery robots operate in various districts of Moscow, in the city of Murino (Leningrad Oblast), and in Innopolis.

We greatly value and support a short cycle from 'idea → concept → testing ground → production,' and we truly have a lot of data and enough GPU power for experiments. Join us!

Our team is responsible for the movement brain of the delivery robot: from scene understanding and dynamics prediction to trajectory generation. Our strategy is to combine the development of transformers with R&D approaches based on an LLM- or DriveGPT-stack and optimize them for the resource constraints of the onboard delivery robot.

You can read an overview article (in Russian) about our delivery robot's architecture on Habr.

What tasks await you

Implementing SOTA approaches You will regularly study the latest approaches, such as Wayformer, PlanTF, MotionLM, DriveGPT4, Diffusion Planner, and adapt them into production-ready solutions.

Improving models with RL You will develop RL-based pipelines (PPO, GRPO, DPO), build reward models, collect datasets in the simulator, and minimize the sim-to-real gap.

Model interpretability You will need to propose ideas for developing model introspection tools—both for debugging complex cases and for obtaining behavior prompting technologies.

More about ML at Yandex — in the channel Yandex for ML

We expect you to

• Have worked with PyTorch
• Have embedded ML solutions into real products
• Have extensive experience with DL, transformers, generative models

Will be a plus

• Have developed and implemented AV ML solutions, are proficient in Imitation and Reinforcement Learning technologies, SLAM, Motion Planning
• Have worked in robotics (familiar with terms like ROS, Jetson, 'sensor set')
• Have participated in ML or programming competitions: ACM ICPC, Codeforces, Topcoder, Kaggle, etc.
• Can write production-ready code in C++