Remember those scenes from the Terminator when the T-800 used to scan its surroundings in order to identify targets or pieces of clothing that might fit? Well, Tesla’s Autopilot system is a lot like that too.
In fact, the EV maker decided to unveil a special landing page dedicated exclusively to Autopilot. There, we can see what they’ve achieved in terms of hardware, neural networks, autonomy algorithms, code foundations and more.
In their own words, Tesla’s take on autonomous tech is this: “We develop and deploy autonomy at scale. We believe that an approach based on advanced AI for vision and planning, supported by efficient use of interference hardware is the only way to achieve a general solution to full self-driving.”
In order to see, Autopilot relies on per-camera networks that analyze raw images, performing semantic segmentation, object detection and monocular depth estimation.
“Our networks learn from the most complicated and diverse scenarios in the world, iteratively sourced from our fleet of nearly 1M vehicles in real time.”
According to Tesla, a full build of Autopilot neural networks features 48 networks that take 70,000 GPU hours to train, boasting an output of 1,000 predictions per every single ‘timestep’.
As the car is moving, autonomy algorithms create a high-fidelity representation of the world, calculating trajectories within that space.
“In order to train the neural networks to predict such representations, algorithmically create accurate and large-scale ground truth data by combining information from the car’s sensors across space and time.”
While you can read a more in-depth description of the program on the landing page by clicking here, we should mention that, as reported by Electrek, the automaker also released an image featuring the Cybertruck in the background of a simulation – not much of an Easter Egg, but still.
Tesla is in the process of hiring new engineers capable of further developing these technologies, so if you fancy yourself clever that way, you might want to fill out the application at the bottom of the aforementioned page.