One of the best things about computers is that they can learn just as much from simulation as they can from so-called “real-world” experience. This means that, given the right simulator, we can teach AI to drive cars without ever putting anyone in danger.
Almost every company with artificial intelligence trains its algorithms for driverless vehicles using simulations. Until now, the simulators themselves were not so interesting. They are primarily physical motors designed to be interpreted by a neural network. But Sony has just unveiled the most popular autonomous driving simulator ever: Gran Turismo Sport.
In case you’re not a gamer: this is not advanced software designed to teach AI, this is a game. And not just any game, but the latest in one of the most beloved racing simulation series in history
Researchers from the University of Zurich and Sony AI Zurich recently published a pre-print paper showing the development of an autonomous agent designed to defeat the best human players in the game.
For the team:
Among the racing games, the Gran Turismo Sport (GTS) is known as an extremely realistic driving simulation, modeling phenomena such as the influence of tire temperature and the car’s current fuel level on traction. Therefore, like real-world racing, the optimal trajectory (ie the trajectory leading to the fastest lap time) for a car in the GTS depends not only on the geometry and properties of the track, but also on different (a priori unknown) physical characteristics and conditions of the car. Due to its similarity to actual driving and the relatively low cost of training in GTS compared to training with actual racing cars, the GTS is also used to cast drivers for racing teams.
In other words: This is a legitimate simulation used by racing teams in the real world to help determine the real abilities of pilots at the expert level. This is a pretty high praise for a video game.
The researchers had a fairly high order to fill. While artificial intelligence systems regularly outperform humans in games such as chess and Go, standard computer-controlled players tend to do poorly against expert human players.
As far as we know, the built-in non-player characters (NPCs) included in modern car racing games are unable to compete with human experts in fair comparisons. For example at the moment The built-in NPC in the Gran Turismo Sport (GTS) loses a total of 11 seconds compared to the fastest human driver and is slower than 83% of all people in one of our reference settings.
Other racing games apparently close the gap to human experts by giving an unfair advantage to the NPC, for example by increasing the engine power of the NPC car; however, this leads to frustration among players who feel cheated.
Instead of cheating or tweaking the rules, the team turned to an aspect of AI called in-depth reinforcement training. This included training the AI to recognize the way forward and to respond in a more humane way.
According to an article by Tech Xplore writer Ingrid Fadela, Yunlung Song, co-author of the team’s research, says:
Unlike classical condition assessment, trajectory planning and optimal control methods, our approach does not rely on human intervention, human expert data or explicit road planning. We found that it can generate trajectories that are qualitatively similar to those selected by the best human players, while surpassing the most famous people tour times in all three of our reference settings, including two different cars on two different runways.
As far as we know, this is the first time a standalone AI beats human experts in Gran Turismo Sport. And although there is currently no artificial intelligence system capable of providing level five autonomy (capable of driving a vehicle without external aids or human assistance), if you absolutely have to drive in an AI-controlled vehicle: you can Choose training in a video game to move the physical limits of speed and control.
You can read the full article here.
So you are interested in AI? Then join our online event, TNW2020where you will hear how artificial intelligence is transforming industries and businesses.
Published on September 15, 2020 – 22:03 UTC