Description
Our team develops control algorithms for Sber's own anthropomorphic robot. The current focus is on whole-body control, training a neural network controller capable of executing any command — from waving hello to dancing and performing acrobatics.
Responsibilities
- Developing robot control algorithms based on training the robot model in a simulator using Reinforcement Learning and Imitation Learning methods
- Implementing and adapting SOTA (State-of-the-Art) papers from the field
- Testing and debugging algorithms on the real robot
- Improving the robot model and learning environment, solving the Sim2Real transfer problem
Requirements
- Deep knowledge of linear algebra and optimal control
- Advanced development skills using PyTorch
- Experience with Reinforcement Learning and Imitation Learning approaches
- Experience working with MuJoCo and Isaac Sim simulators
- Experience in the field of robotics, with real robots, is desirable
- Having publications at ICRA/IROS/ICML/NeurIPS and other Robotics and DL conferences and journals is a plus
Conditions
- On-site work format in Moscow
- Annual salary review and yearly bonus
- Comprehensive voluntary health insurance (VHI) and preferential family insurance
- Sber's unique training system for professional and career development
- Favorable mortgage program for employees
- Free SberPrime+ subscription, discounts on partner company products
- Referral bonus for recommending friends to join the Sber team
- Corporate pension program.
