Tesla engineers its own computer for ‘full self-driving’
Autonomous Vehicle Technology announces 2020 ACES Award Winner in Autonomy | Systems category
In April at its first Autonomy Day investor event, Tesla raised the bar again for automakers trying to compete with the new automaker on vehicle autonomy. It revealed details of its new FSD (Full Self Driving) computer hardware and software, which was put in production a few weeks before. The company took over full control of development in an effort to put some space between it and other competitors. Some of the top goals of the development were to keep total power below 100 W performance of at least 50 TOPS (trillion operations per second)—all while lowering its per-unit costs. Compared with the HW 2.5 it replaces, the FSD computer is 80% of the cost. The new system can process 2300 vs. 110 frame/sec. Power consumption rises from 47 to 72 W, staying comfortable under the 100 W goal. It uses two of its new AI (artificial intelligence) chips for redundancy, each equipped with a CPU, GPU, and deep-learning accelerators. It delivers 144 TOPS (trillion operations per second), collecting data from a range of camera, radar, and ultrasonic sensors, and is power by deep neural-network algorithms. Tesla also has an advantage on training data for its deep neural networks. Its fleet of about 500,000 vehicles on the road with the FSD means it is driving around 15 million miles each day, and Tesla’s fleet is growing by about 5000 cars per week. By mid-2020, the company says it will have 1 million cars on the road that could be switched onto a robotaxi network to improve its car’s value to consumers. Another differentiator from traditional OEMs is that Tesla plans to retrofit the new computer for customers having previous hardware but have purchased its advertised FSD functional package. Every other automaker will be challenged to deliver this level of performance and service.
Click to view AVT’s 2020 ACES Award Winners best in class of