RL Engineer for the Humanoid Robot Locomotion Team

Solving the locomotion problem is one of the key challenges in modern robotics. We create and train models that teach robots to walk, maintain balance, and perform complex motor tasks in simulation and in the real world.

You will work with RL agents in Isaac Lab and MuJoCo, develop physical scenes: from stairs to rough terrain, adapt modern approaches like Residual RL and Diffusion Policy to real-world motion tasks.

The goal is to teach the robot to move naturally, stably, and safely.

What tasks await you

Training RL agents for locomotion You will create and train walking, balancing, and complex motor skill policies in Isaac Lab and MuJoCo.

Research and application of modern RL methods You will adapt various ideas from scientific papers to locomotion tasks: from Residual RL to transformer-control.

Creation and complexity of simulation environments You will need to design physical scenes on which robots will learn to move: stairs, uneven surfaces, obstacles.

Analysis and improvement of agent behavior You will develop metrics, validate reward functions, and search for non-obvious dependencies and growth points in model behavior.

Implementation of developments on real robots You will transfer trained policies to real platforms and observe how your models start moving in the physical world.

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

We expect that you

• Have trained RL models and implemented solutions in production or research pipelines
• Confidently program in Python and work with PyTorch
• Understand robot kinematics and dynamics
• Are well-versed in modern RL approaches such as PPO, SAC, TD3, etc.
• Can formulate hypotheses and set up experiments to test them
• Read and implement ideas from scientific papers from ICRA, CoRL, NeurIPS, RSS, and others
• Have experience working with Isaac Lab, Isaac Gym, and MuJoCo

Will be a plus

• You know C++ or CUDA, have experience with physics simulation and computation optimization
• Understand the Early Experience and Diffusion Policy approaches