Autonomous vehicles must always know their precise position in space. One of the most reliable methods for solving this task is localization using a lidar map: a three-dimensional reconstruction of the world as a point cloud.

The lidar map development team is involved in:

• Mechanisms for automated construction and updating of the lidar map
• Algorithms for global point cloud optimization underlying these mechanisms

What tasks await you

Development of optimization algorithms You will improve optimization algorithms on a factor graph. Develop metrics for matching between point clouds. Add additional information (segmentation, local cloud properties) to the matching. Improve global optimization based on both past and future data.

Detection of map obsolescence and its automatic rebuilding It will be necessary to determine the irrelevance of the lidar map based on the fleet's driving data and develop methods for automatic map rebuilding using this same data.

Automated verification of the quality of the built map You will create methods to automatically verify the map: find unsuccessful optimization results and assess localization quality.

We expect you to

• Have experience working with point clouds
• Have solved optimization problems
• Possess a strong mathematical background and be ready to dive into the world of state-of-the-art optimization algorithms
• Be proficient in C++ and Python
• Know classic algorithms and data structures

Will be a plus

• Have solved SLAM tasks or worked with optimization on factor graphs
• Know what ICP and NDT are
• Have experience with Ceres, g2o, or GTSAM
• Have used cloud data orchestration tools (Yandex Nirvana, Airflow, Dagster)