Kirill Brodt, an MTS AI employee, came third in the international competition, having developed a unique model for predicting the trajectory of road users.
The MTS AI employee ranked 3rd in the international AI competition
Kirill Brodt, an MTS AI employee, took a prize at the Waymo Motion Prediction Challenge. The Waymo Challenge is one of the most prestigious AI and ML challenges in the world. 19 teams from different countries developed the most accurate models predicting the behavior of road users.
The competition was attended by young scientists developing projects in the field of CV and predictive analytics. The teams faced the challenge of observing for one second and predicting the actions of road users over the next eight seconds. For an unmanned vehicle, this time is enough to instantly analyze the traffic situation and react correctly, thereby avoiding a collision or accident.
Kirill’s team managed to predict the trajectories of road users with an accuracy of several tens of centimeters relative to their actual spatial positioning. Kirill worked on this solution as part of the international team comprised of Stepan Konev (Skoltech) and Artem Sanakoev (University of Heidelberg).
According to Kirill, the complexity of the task lay in working with data consisting of a large table with numerous parameters: the position of road objects, agents (cars, pedestrians, cyclists), speeds, angular speeds and the directions of the agents. Training the model to predict trajectories eight seconds ahead using standard CV methods took at least three days on one video card. According to the competition results, Kirill and his team came third.
Within the framework of cooperation between Novosibirsk State University (NSU) and MTS AI, Kirill Brodt is engaged in projects exploring 3D spatial reconstruction and preparing NSU students for machine-learning competitions. Next year, he plans to become a teacher of a practical course for AI students.
In June, at SPIEF, MTS AI and the Novosibirsk Regional Government signed an agreement on the creation of an innovative AI tech cluster for cooperation between NSU scientists, inventors, MTS AI experts and startups.
Aleksandr Khanin, Director of MTS AI, noted that the team’s achievement at the international competition is the result of efficient work with the scientific community, which is useful for scientists, business and the industry in general. The solution developed by Kirill and his team will form the basis for creating a functioning mechanism for predicting the behavior of road users in terms of a self-driving car. The unusual approach proposed by the team to solving the task will be added to global experience in this area.
