Senior DL Perception Engineer

Delivery robots are a step into the future of logistics and autonomous technologies. We are creating a system where robots will be able to move autonomously around the city, interact with people and infrastructure, delivering orders quickly and safely.

You will face exciting engineering challenges: from computer vision and navigation to distributed systems and real-time machine learning. We are looking for talented specialists who want to work with cutting-edge technologies, see the results of their code in action, and influence how robots become part of everyday life.

What tasks await you

Improving Recognition Algorithms You will be responsible for developing and improving neural network algorithms for computer vision in the fields of 2D, 3D recognition and tracking. You will formulate hypotheses, conduct experiments using accumulated data, and analyze results using existing and new quality metrics. Your goal is to create and implement architectures capable of accurately and compactly describing the surrounding world of a delivery robot.

Development and improvement of the Data Engine The larger and more diverse the data, the better the models become. We are developing an efficient Data Engine that automates data collection, selection, and annotation, reducing costs and accelerating the improvement cycle. You will work on tools and algorithms that improve the quality and speed of annotation (including with the involvement of external annotators), as well as ensure maximum return from each new data example.

Development of experimental technologies We are not limited to just improving existing solutions — we are exploring new approaches to robot perception and world understanding. You will participate in researching foundational models, studying end-to-end approaches, and integrating new architectures into the robot's perception system. We expect you to actively propose and test bold ideas that can radically increase the efficiency and reliability of our robots.

Model Optimization All our models run directly on the robot's board — in real-time and with limited resources. Therefore, we expect that any new solutions will not only be accurate but also performant enough to execute on the robot. You will solve optimization tasks at various levels: from architectural decisions and efficient code writing to developing custom operations for TensorRT and other inference acceleration tools.

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

We expect that you

• Are highly proficient in the world of applied computer vision: have a solid knowledge of theory and methods for working with CV tasks, have successfully applied them and achieved practical results
• Are well-versed in linear algebra, probability theory, and projective geometry
• Have integrated CV and ML solutions into products
• Have worked with PyTorch, understand the intricacies of parallel computing well, write efficient GPU code
• Program in Python or C++
• Can handle complex tasks under tight deadlines
• Can explain your decisions and defend them in discussions
• Can work in a team
• Are responsible and organized

Will be a plus

• Have solved object detection and tracking tasks in 2D or 3D
• Have worked with sensors other than cameras: lidars, radars
• Are familiar with TensorRT, neural network acceleration, quantization, and have experience writing custom operations
• Have participated in artificial intelligence competitions or competitive programming