A team of AI experts based in the UK have managed to teach an autonomous car to stay in a lane and perform basic self-driving functions in just about 20 minutes.
Goes to show just how advanced AI has become, as well as its potential to have human-like intelligence in its respective field. Newly developed neural networks have already been beating humans in complex tasks that require cognitive decision-making and intuition – something machines have lacked in the past.
Wayve, a tech company run by Cambridge University’s Engineering Department members, unveiled its “reinforcement learning” algorithm back in June and its capability to learn new things, such as driving, by working with a human safety driver.
This is not the first time humans were beaten at their own game by AI software, neural networks that work on similar principles as Wayve’s reinforcement learning have learned to beat Chess and Go grandmasters.
The team uploaded a video on YouTube showing how they taught their reinforcement learning algorithm to drive a car. Check it out below,
First Time Ever in a Self Driving Car
In their video, the company emphasized the fact that their new algorithm is “the first example of reinforcement learning onboard an autonomous car.”
It works differently compared to most recent iterations and uses minimal hardware to accomplish the task. Developers did not install a heap of sensors and other components, instead, a single GPU has been installed on-board that keeps learning from its mistakes.
The AI takes random turns in the beginning but slowly stabilizes as the driver keeps correcting its mistakes. The modified Renault car does not need “massive models, fancy sensors and endless data” but drives itself autonomously via “a clever training process that learns rapidly and efficiently.”
As you’d expect, this automatically reduces the costs and might be the cheapest self-driving vehicle yet. It will soon get more advanced, and add more features such as “dealing with traffic lights, roundabouts [and] intersections.”
Check out Wayve’s blogpost for more details.