One of the best things about computers is that they can learn just as much from simulation as they can from what is known as “real world” experience. That said, with the right simulator, we can teach the AI how to drive a car without ever putting a single person at risk.
Almost every AI company trains its driverless vehicle algorithms with the help of simulations. So far, the simulators themselves haven’t been that interesting. Most of them are just physics engines that are supposed to be interpreted by a neural network. But Sony just unveiled the most popular autonomous driving simulator of all time: Gran Turismo Sport.
In case you are not a gamer, this is not advanced software for training AIs, it is a game. And not just any game, but the latest in one of the most popular racing simulation series in history
Researchers from the University of Zurich and Sony AI Zurich recently published a pre-print paper presenting the development of an autonomous agent to beat the game’s best human players.
Among the racing games, Gran Turismo Sport (GTS) is known as an extremely realistic driving simulation that models phenomena such as the influence of tire temperature and the current fuel level of a car on traction. Therefore, the optimal trajectory (i.e. the trajectory leading to the fastest lap time) for a car in GTS, similar to real races, depends not only on the geometry and the properties of the track, but also on different ones (unknown a priori). physical properties and states of the car. Due to its similarity to real driving and the relatively low price for training in GTS compared to training with actual racing cars, GTS is also used to cast drivers for racing teams.
In other words, it is a legitimate simulation used by real-world racing teams to assess the skills of drivers at an expert level. That’s pretty kudos for a video game.
The researchers had a pretty big job to do. While AI systems routinely perform better than humans in games like chess and go, standard computer-controlled racing drivers tend to hold their own against experienced human players.
The researchers write:
To the best of our knowledge, the built-in non-player characters (NPC) are included in modern car racing games are unable to compete fairly with human experts. For example the current one The NPC built into 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 the NPC an unfair advantage, for example by increasing the engine power of the NPC’s car. However, this creates frustration for human gamers who feel they have been betrayed.
Rather than cheat or tweak the rules, the team turned to a facet of AI called deep learning. This included training the AI to see the path ahead and to respond more humanly.
According to an article by Tech Xplore writer Ingrid Fadella, Yunlung Song, co-author of the team’s research paper, said:
In contrast to classic condition estimation, flight path planning and optimal control methods, our approach is not based on human intervention, human expert data or explicit path planning. We have found that it can produce trajectories that are qualitatively similar to those of the best human players while outperforming the most famous human lap times in all three of our reference settings, including two different cars on two different tracks.
To the best of our knowledge, this is the first time autonomous auto AI has beat human experts in the Gran Turismo Sport. And while there is currently no artificial intelligence system capable of achieving level five autonomy (being able to drive a vehicle without external aids or human assistance), if you absolutely must be in an AI controlled vehicle you can have to drive, including the trained to choose in a video game by exceeding the physical limits of speed and control.
You can read the whole paper here.
So you are interested in AI? Then Take part in our online event TNW2020Here you can find out how artificial intelligence is changing industries and companies.
Published on September 15, 2020 – 22:03 UTC