Tesla is committed to providing Full Self-Driving functionality to all of its customers. To do this, the company is constantly developing improved hardware and software.
The company developed and designed Full Self-Driving (FSD) Computer from the ground up for fast processing of neural networks. At the moment, Tesla cars are not completely autonomous and require active control by the driver, and the FSD computer is able to provide intelligent performance and control, providing a new level of safety and autonomy.
Unlike lidar, which uses infrared radiation (which is well absorbed by water vapor), Tesla cars use 8 standard cameras with which they train a computer, radar and ultrasonic sensors. Of course, they don’t see what a person sees and cannot recognize objects just as easily, therefore Tesla is developing teaching methods and systems.
To reliably distinguish complex objects, the computer needs data on how a complex object can look from any point of view. Tesla cannot teach a computer to think, but it is able to give its computers enough images so that they can distinguish complex objects in any conceivable environment.
In March, Tesla published patent 'System and method for obtaining training data', in which Tesla’s head of AI and Autopilot software Andrej Karpathy describes about improvements in the process of computer training.
Yesterday Tesla published a new patent, 'System and methods for training machine models with augmented data', in which DeepScale startup members (Matthew Cooper, Paras Jain, and Harsimran Singh Sidhu) talk about systems and methods for learning data in a machine learning environment and, in particular, about increase training data by including additional data, such as sensor characteristics, in the training data set.
This invention involves training a neural network using data from several different sensors observing scenes, for example, eight cameras in a Tesla Autopilot sensor array.
As we can see, the California automaker is actively striving for Full Self-Driving which does not require driver intervention. According to Tesla CEO Elon Musk, the Robotaxi fleet should appear by the end of 2020 (if regulatory authorities allow this), which means that functionally, all the company's cars should be able to drive independently by the end of this year.