ML Engineer for Delivery Robot Development Team

In the Autonomous Transport division, we create and scale solutions that operate in the real physical world and are integrated into complex logistics and operational processes.

Within the Robot Delivery division, a new technical direction is being launched, enabling a qualitative expansion of the capabilities and functionality of delivery robots. Expanding functionality requires solving substantially new tasks and developing new technological approaches in almost all areas of autonomous transport: architecture, key algorithms for perception, localization, and planning, as well as in logistics and operations.

We are looking for a strong developer to join the team that will go from the first prototypes to a mature product.

What tasks await you

Development and adaptation of new approaches New methods of machine learning in autonomous driving and robotics emerge every day. They need to be adapted for the robot under development, new approaches must be created based on them, and thorough training must be conducted, including data selection, data cleaning, setting up training pipelines, and analyzing results.

Adapting algorithms to computational constraints We develop useful robots that must perform their function and generate profit for the company. Therefore, we are limited in computing power and sensor set. At the same time, we have accumulated a significant amount of data and models that need to be placed on our computing platform. Fitting all models into limited memory and performance is a creative task. We use various methods of model quantization, architecture optimization, and knowledge transfer (Transfer Learning) to solve it.

Development of the pipeline and ML models Developing a robot requires creating and evolving not only internal behavior models (localization, navigation, perception) but also models of the external environment. You will study the existing vector map-building pipeline and improve it, develop and train ML models for extracting and vectorizing elements of road infrastructure. You will need to research and implement new approaches from scientific papers (CV, ML, 3D, multimodal learning), propose and implement your own ideas and algorithms. You will work with real autonomous transport data (sensors, labeling, noise, scale), participate in the full R&D cycle — from idea and experiments to production solutions — and analyze model quality, influencing safety and reliability metrics.

Learn more about ML at Yandex — in the channel Yandex for ML

We expect you to

• Have developed ML algorithms from data collection to product operation
• Have a broad knowledge of ML methods, especially in the fields of robotics (localization, route planning, recognition)
• Have experience in profiling and adapting neural network algorithms for platforms with limited computing power (quantization, pruning, Transfer Learning)
• Be able to work in a team
• Be responsible and organized

Will be a plus

• Knowledge of classical ML methods: ensemble methods (boosting, bagging), support and relevance vector machines (SVM, RVM), methods of conditional and convex optimization