Roadbound Robots: Toronto start-up develops self-teaching driving AI

Markham, Ontario — Afraid of artificial intelligence (AI) capable of replacing people? Well, someday, this AI could replace the driver’s seat altogether.

Toronto tech start-up, Waabi has created a realistic virtual environment where AIs teaches itself how to drive. This encompasses a huge range of factors from object reconstruction, camera and LiDAR input along with real-time scenarios for the AI to train in.

Unlike other machine learning algorithms (think of YouTube’s recommendations), autonomous driving systems must be able to interpret visual data in real-time and ultimately drive as safely, or safer than a human. Crashing a vehicle has more consequences than showing you a video from the weird parts of the internet.

Closed-loop simulations use incoming data to refine the AI’s output, creating better results without human input. Infographic by Waabi.

Like other autonomous learning systems, Waabi World does not use human input beyond the initial coding. Instead, it replicates how the AI sees the world through its sensors, allowing the AI to tune itself. Using a more efficient physics simulator that accounts for fog, exhaust or reflecting light, Waabi World can conduct more tests than its counterparts, developing more sophisticated, and safer driving skills.

Additionally, the world itself ‘challenges’ the AI by constantly identifying its weaknesses and creating adversarial scenarios. Each failure identified is a chance for the AI to improve its driving skills, and another chance to prevent accidents in the real world. In theory, this could lead to AIs that drive safer than any human ever could.

Collision Repair, for one, will welcome our new automated overlords–will you?


Sign-up for the Collision Repair daily e-zine and never miss a story –  SUBSCRIBE NOW FOR FREE!

Related Posts

One Response

Leave a Reply

Your email address will not be published. Required fields are marked *