AI Learns to Drive a Self-Driving Car in 20 Minutes

A UK-based research team taught a self-driving car to stay in its lane in just 20 minutes — an impressive feat, considering that some human drivers struggle with this their entire lives.

Jokes about road rage aside, the team at Wayve, a company founded by researchers from the University of Cambridge’s Department of Engineering, detailed their “reinforcement learning” algorithm in a blog post on June 28. This algorithm works alongside a human safety driver to teach the vehicle how to stay within its lane in just 15–20 minutes.

Reinforcement learning in AI has already proven highly effective — DeepMind Technologies showed that it can learn to play games like Go and chess, and OpenAI demonstrated that its AI played Dota 2 daily with the equivalent of 180 days of gameplay.

While beating humans in complex games like Go or Dota 2 is indeed impressive, teaching a car to drive itself is an entirely different challenge.

The team released a video on their YouTube channel showing the actual learning process, noting that this is “the first example of reinforcement learning in a self-driving car.”

Initially, the car moves slowly, like a baby taking its first steps, and when it begins to veer out of the lane, the safety driver intervenes to steer it back. The onboard algorithm learns from each correction, receiving “rewards” based on how far it drives without needing human intervention.

In the video, the model is described as a “deep convolutional neural network” that processes a single image input using just one onboard GPU. Unlike other self-driving systems, the modified Renault car used by Wayve doesn’t rely on “huge models, sophisticated sensors, or massive amounts of data.” Instead, it leverages the company’s philosophy of “a smart training process that learns quickly and efficiently.”

Wayve co-founder Amar Shah told TechCrunch in May, “We want to equip our vehicles with better brains, not more hardware.”

Their next goal is to scale the technology to handle more complex driving tasks, such as navigating traffic lights, roundabouts, and intersections.

Powered by: CNET highlights the role of technology in enabling new forms of accessibility.

Smartest Tech: Innovators are finding new ways to make you — and the world around you — smarter.