Senior Developer for the Delivery Robot ML Planner Team (RL)

Every day, Yandex delivery robots transport food and parcels along the streets of cities in Russia.

Our task is to develop advanced planning and trajectory forecasting models that are trained on billions of simulation steps and successfully deployed to production.

Our team is responsible for creating and improving the RL planner, which makes millions of decisions daily on roads and sidewalks. We actively experiment with cutting-edge Reinforcement Learning approaches and transformer architectures, utilize a powerful GPU simulator, and continuously seek new methods for training models and improving their quality.

Want to create algorithms that help delivery robots move safely and efficiently in the real world? Join us so your code controls the movement of the robots you see on the streets!

What tasks await you

Research and integration of new approaches for RL planner training You will need to explore modern RL methods and their modifications (PPO, GRPO, TD-λ, and others), prototype new algorithms, configure reward functions, and training scenarios. Your task is to turn ideas from papers and experiments into a robust production pipeline.

Development of transformer-based planning model architectures You will be designing and implementing new modules into models based on Wayformer and MotionLM approaches. You will be responsible for model quality, training stability, and model behavior interpretability.

Development and expansion of simulation scenarios As part of the team, you will devise new complex scenes for the GPU simulator. The goal is to close the sim-to-real gap and make the planner as robust as possible to unexpected situations in the city.

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

We expect you to have

• Proficient knowledge of Python and PyTorch
• Understanding of the fundamentals of Reinforcement Learning and transformer models
• Ready to study complex approaches and solve non-trivial tasks

Will be a plus

• Experience with RL, Motion Planning, or transformers
• Knowledge in robotics and simulation
• Experience with C++, ROS, TRT, Cuda