Tesla is determined to win the autonomous race without lidar.
Tesla did not comment on the reported purchase, though CNBC also reported DeepScale CEO Forrest Iandola made a curious announcement on LinkedIn. On the social media network, Iandola confirmed he joined Tesla as a senior staff machine learning scientist. CNBC’s sources familiar with the deal reported back saying it wasn’t a single hire and that Tesla has, in fact, purchased the startup outright.
DeepScale’s approach to autonomy fits the bigger picture Musk has promoted for a few years now. Rather than relying on lidar, Musk has consistently believed cameras, radar and ultrasonic sensors will make up a robust system without other hardware. Powering it all is a new artificial intelligence chip Tesla developed in-house. The chip, detailed this past April, uses minimal power for operation and takes in an absolute massive load of information from the hardware package.
Meanwhile, Tesla also detailed a neural network that will help bring autonomy to life. This strategy also appears to sync well with the reported DeepScale acquisition. This neural network pulls video and images from Tesla vehicles in the real world and learns how to read roads more efficiently and accurately. Human annotation and predictive behavior (more machine learning) power the process. Andrej Karpathy, Tesla’s head of AI, earlier stated that the neural network and other hardware will be vastly superior to lidar.
If all goes according to plan, Tesla wants to have a 1 million-strong test network of robotaxis on the roads by the end of 2020. DeepScale could, perhaps, help it get there.