NVIDIA has learned that researchers at MIT are developing an all-in-one deep neural network (DNN) to power self-driving cars and replace systems composed of multiple networks.
It is reported that the research published at this year’s COMPUTEX used NVIDIA DRIVE AGX Pegasus to run the network in the vehicle, which can process lidar data efficiently and in real time.
Self-driving car sensors generate vast amounts of data. If a fleet of 50 vehicles travels 6 hours a day, that would generate about 1.6PB of sensor data per day.
In the paper, the MIT team details how to use an all-in-one DNN to try out new autonomous driving strategies, starting with real-time lidar sensor data processing tasks. Using NVIDIA DRIVE AGX Pegasus, the team further accelerated lidar computation to achieve or exceed this goal, running 15 times faster than current state-of-the-art systems.
The reporter learned that NVIDIA said that NVIDIA DRIVE AGX Pegasus is an AI supercomputing platform designed for L4 and L5 autonomous driving systems. It combines the performance of two NVIDIA Xavier SoCs and two NVIDIA Turing architecture GPUs to achieve 320 teraflops of performance.
Source: C114 Communication Network
Author: lazy cat