Ivan Chistiakov Email and Phone Number

Development of unified framework for always-on navigation solution.• Combining existing approaches into one project and refactoring the code.• Designing data collection protocol to cover complex scenarios with device position / activity switches.• Training lightweight NN models for both smartphones and wearable devices, integration with EKF is considered.Research on indoor / outdoor swimming workout analysis. • State-of-the-art swimming style classification model is shipped for wearables.• Mileage error is decreased by XX% in the outdoor case.

• New AI-PDR algorithm is proposed through close cooperation with a university.• Positioning accuracy is improved by XX%, which solves the pain point of trajectory drift in harsh scenarios such as outdoor urban canyons, high-rise buildings, and double-sided shelters.• The algorithm is released in Mate 50 & Mate X5 series.

• Giving lectures on Optimal control theory.• Continuing work on PhD thesis.

During this stage, we published the RuDaCoP dataset for development of smartphone-based pedestrian navigation algorithms. The data are accompanied by the high accuracy ground truth collected with two foot-mounted inertial measurement units.Personal contributions:• I played a key role in algorithm development for ground-truth preparation.• The code was optimized. In addition, the original MATLAB scripts were transformed into C code with a Python interface for higher performance.• The results were compared with RTK solution to make sure of the accuracy.Results:• We managed to publish a large dataset which consists of about 1200 sets of inertial measurements from sensors of several smartphones.• These data were used internally for development of deep learning algorithms in several key projects.

During the internship, I was working on accurate reconstruction of pedestrian trajectories using inexpensive foot-mounted inertial navigation systems (IMU).The solution is based on the following ideas:• Kalman filtering is applied to make use of zero velocity observations when a foot touches the ground.• IMU devices mounted on both legs are synchronized to provide additional information to each other.Results:• Final positioning error is less than 10% of the travelled distance.• The method was later presented on the ICINS-2019 conference and in the Russian Technological Journal.• The algorithm was used for dataset collection to tackle different problems of pedestrian navigation.