In Minsk at ATOM GLOBAL AI EXPO showed a drift car controlled by a neural network
Demonstrative drive of the drift car, which was controlled by a neural network, took place at the site near the Minsk International Exhibition Center in Minsk.
A demonstration drive of the drift car, which was controlled by a neural network, took place at the site near the Minsk International Exhibition Center as part of the ATOM GLOBAL AI EXPO forum on October 18. There was no driver in the car – all maneuvers were performed by an artificial intelligence system. Details - in the material of the correspondent of the agency "Minsk-Novosti".
The race started at 12:00. The developers emphasize: the car is not an autopilot, namely a neural network.
- Our project is the creation of a drift car controlled by a neural network. Not autopilot, but neural networks. We solve the problem of machine learning: we have created a drive-by-wire control system – it works on the principle of electronic signals – we have integrated it into a sports car, developed a model of a neural network and a pipeline – the whole process of its training – which allows performing drift sports elements by training with reinforcement, said project manager Konstantin Nikiforov.
He explained that in order for the neural network to learn to drift, it is first trained in a virtual environment, where it works out maneuvers, and then transferred skills to a real machine.
On the site, the car performed sports drift elements: sharp turns, went into the skid and made bundles of elements. There was no driver inside - all control went through the neural network.
- We see the virtual learning environment, the management system, the learning process and how the management system responds to it. And at the end is a car in which all these technologies are connected. Today we kicked him out into the street, ran our algorithm and watched him drift, he said.
The project, according to the developers, reassembled from scratch a little more than a year ago.
Photo by author
