ML Developer for the Reinforcement Learning (RL) Group

ML Developer for the Reinforcement Learning (RL) Group

Solving the motion planning task is one of the biggest challenges in the autonomous technology industry. We are experimenting with large generative models to predict a vehicle's trajectory. Such a trajectory must ensure safe interaction with other traffic participants, be comfortable for passengers, and help the vehicle reach the final destination.

To improve the model's performance in new traffic situations, we are looking for ways to enhance it: we research new RL training algorithms and work with data. To improve driving quality in non-standard situations, we create synthetic complex traffic scenarios and include them in the model's training.

You will be tasked with developing the algorithm so that autonomous vehicles perform reliably in both city and highway environments.

What tasks await you

Improving RL Models You will train transformers to extract information about the interaction of traffic participants with the vehicle's trajectory, test and implement modern approaches from scientific papers on autonomous driving, and experiment with architectures, rewards, and training methods.

Research on Synthetic Data You will experiment with reinforcement learning methods, including the use of simulators, and explore the possibilities of adding synthetic data to such training.

Implementing Developments into Vehicles You will implement your developments into the vehicle's motion pipeline and observe how your work improves the quality of the technology.

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

We expect that you

• Have trained neural network models and implemented DL solutions in production
• Are proficient in the DL field and quickly grasp new methods
• Program in Python and have worked with PyTorch
• Know classical data structures and algorithms
• Can formulate hypotheses and design experiments to test them
• Read ML articles and keep up with developments in this field

Will be a plus

• Have implemented reinforcement learning methods in production
• Have published in proceedings of high-level scientific conferences
• Have worked with large volumes of data
• Have participated in ML or competitive programming competitions
• Are passionate about cars or autonomous vehicles