Accelerating Machine Learning

Paving the way for efficient Deep Learning with NVIDIA GPUs.

- Ian Chee

Challenge

Perception is key in every self-driving car system. With the earlier generations of self-driving cars, the primary technology revolves around LiDAR (Light Detection and Ranging), which is an adaptation of radar technology to use light pulses instead of radio waves.

This is, while possible, very expensive to implement. LiDAR equipment consists of a 3D laser range finder, which generates enormous amounts of data to be processed. This increases costs on two fronts – the visual input device, and the hardware required to process the information from said device.

There is also the problem of predicting the behavior of others. While LiDAR does well in detection, there is no element in place dictating its reaction to a pedestrian walking into its path, or another driver moving erratically on the road.