Self-driving cars have a lot of hurdles in front of them. The biggest one will and is; how do the companies developing the self-driving tech test their AI? AImotive understands driving millions of miles on the real road won’t teach the self-driving AI to account for every possible scenario. Yes, Tesla is collecting data from the actual road in their cars, but freak accidents happen. AImotive built a simulator where they could teach their AI to react to unseen accidents and possibly test the AI’s viability on the open road.
The Deep Drive project is taking a similar approach, but they’re using an unconventional simulator – Grant Theft Auto V. That’s right. They’re using the open world video game where you can drive, kill, and escape the police. They’re opening for testing of self-driving AI’s in real(ish) world conditions. Researchers will have a basis to test their AI before they take the technology to the road or a more advanced simulator.
The DeepDive team claims:
The high fidelity and vast open world provided by GTAV also presents the most complex virtual RL environment to date for testing sensorimotor AI. This allows a new level of testing for safety in AI as existing environments don’t offer the same opportunities for reward hacking, distributional shift, and negative side effects. Finally, developing self-driving cars out in the open provides a level of transparency not usually seen in AI applications and in an area where it is crucial to have visibility into both the safety and correctness of the system.
Hopefully, the AI’s learning in Grand Theft Auto doesn’t pick up the driving habits many players have. With the technology, more self-driving AI’s will get to test their skills in a safe space. More researchers will teach their AI’s to react in smart and calm ways when the AI is thrown into a situation it’s never experienced before.
Tell us in the comments below what you think about AI’s using simulators to learn how to drive, whether it’s in Grand Theft Auto or like sims AImotive has built.