The driving process in a race is complicated for the most specialized drivers. This with the AI ??has been resolved as they can learn to complete a lap as fast as possible. However, everything gets complicated when it is not the only vehicle in the race.
Well, the results obtained by the researchers are given after virtually testing an artificial intelligence program against the best competitors in the video game Gran Turismo Sport.
“In most games, the environment defines the rules and protects users from each other, but in racing, the cars are very close to each other, and there is a very refined sense of etiquette that has to be be learned and deployed by AI agents,” says Peter Wurman, director of Sony AI America.
“To win, they have to be respectful of their opponents, but they also have to preserve their own driving lines and make sure they don’t give way,” he continues.
Gran Turismo 7 is delayed until 2022 due to the pandemic
To achieve this, Sony AI researchers used a technique called deep reinforcement learning. The AI ??was rewarded for certain behaviors, such as staying on the track, maintaining control of the vehicle, and respecting racing etiquette.
Once this was established, they ran the first test (July 2021), pitting their AI, called GT Sophy, against human Gran Turismo champions. In the first phase, the latter achieved the highest overall team score.
However,in the second test, carried out in October of the same year, the AI ??prevailed. He outperformed his human rivals both individually and as a team, posting the fastest lap times.
“In racing, as in many sports, it’s all about getting as close as possible to the perfect lap , but you can never get to it. With Sophy, it was crazy to see something that was the perfect lap. There was no way to go faster,” explains Emily Jones, who was a world finalist in the Gran Turismo Championship 2020 .
Artificial Intelligence will be the protagonist. The role it will have in this 2022
The Sony team, despite this victory, is not sitting idly by and is now further developing the AI. “One of the things we’re looking at is: can we train a single policy that can race any car on any of the tracks in the game?” Wurman asks.
Apart from this, they continue to study how to tune the performance of the AI ??so that it can be a challenging but not invincible opponent, even for less skilled players than the champions who have tested the AI ??so far.
To improve it, all these types of tests are useful, but if you want to incorporate it into the final versions it can be quite frustrating.