|Algorithmic patterns and codes / Photo by: Vintage Tone via Shutterstock|
Intelligent algorithms are the key to improving the performance of self-driving technologies, according to two Ph.D. holders. The pair who founded the startup company, Wayve, managed to teach a non-autopilot Renault Twizy to follow a lane for 20 minutes.
Dr. Amar Shah and Dr. Alex Kendall from Cambridge University started Wayve and their approach in developing autopilot systems involves highly intelligent algorithms. Their approach does not include the use of too many maps, rules, and sensors.
"The missing piece of the self-driving puzzle is intelligent algorithms, not more sensors, rules, and maps. Humans have a fascinating ability to perform complex tasks in the real world because our brains allow us to learn quickly and transfer knowledge across our many experiences. We want to give our vehicles better brains, not more hardware," explained Dr. Shah, CEO and co-founder of Wayve.
To prove it can work, they took a Renault Twizy, a two-seat electric car, and attached one camera at the front area, then tweaked the vehicle so it can operate on its own, such as accelerating, braking, and steering. Meanwhile, a graphics processing unit capable of analyzing data in real-time has been connected to the car.
Once the vehicle was set, they ran a learning program that consisted of data based on experimentation, evaluation, and optimization associated with self-driving in the car’s computer system.
On a narrow, gently curving lane, a human driver sat in the driver’s seat of the Twizy but allowed the car to traverse the lane in autopilot. The driver did not issue any order to the computer system of the car, and simply watched how it handled everything.
After recognizing the road, the car accelerated and stopped several times as the algorithm corrected the autopilot system. If the algorithm detects correct driving, the system is “rewarded” while a wrong action puts the vehicle on hold and gets “penalized” by the system for its mistakes. For 20 minutes, the car managed to follow a lane more smoothly.
The algorithm used in the vehicle follows the similar principle of DeepMind, a deep reinforcement learning method used in many computer games like Go and Chess.