In early March, Tesla CEO Elon Musk confirmed that a new Full Self-Driving (FSD) Beta version would widely release in April. As per multiple patents and confirmation by Musk, this new FSD Beta will use pure vision.
Pure vision FSD would allow a vehicle to detect and interpret the distance to its surroundings through image data and machine learning. Instead of traditional radar, FSD Beta utilizes real-work AI through the establishment of two neural networks.
The first network determines the distance of objects from live images produced by the car’s cameras. The second network is the training network to help the first one learn through machine learning, reports Teslarati.
FSD Beta aims to find the optimal configuration of sensors to limit the bandwidth strain while still collecting the correct data.
The patents also include a way to automatically label vision data. This process is traditionally is one of the most time-consuming parts of machine-learning. With this new patent objects are automatically uploaded to the training network without any human intervention.
Early user-experience on FSD Beta has been positive with its limited release. It is early, but we are excited to see where this goes as FSD Beta pure vision gets tested in the real-world.