Low cost & power CNN/Deep learning solution for Automated Driving

Mihir Mody, Kumar Desappan, Pramod Swami, Manu Mathew, Soyeb Nagori
Texas Instruments, Inc


Abstract

Automated driving functions, like highway driving and parking assist, are increasing getting deployed in high-end cars with the ultimate goal of realizing self-driving car. This talk starts with an introduction to automated driving. Deep learning techniques like convolution neural network (CNN) are one of the key enablers to achieve many subfunctions for automated driving. For mass-market deployment, the embedded solution is required to address the right cost and performance envelope along with security and safety. Also, given solution shall be scalable to adopt various network topologies. In the case of automated driving, one of the key functionality is “finding drivable free space”, which is addressed using deep learning techniques like CNN. These CNN networks pose huge computing requirements in terms of hundreds of GOPS/TOPS (giga or tera operations per second), which seems beyond the capability of today’s embedded SoC. This talk covers various tradeoff parameters when tailoring for low cost & power solution. Finally, this talk concludes by demoing the results of a CNN network running in real time on TDA2X SoC, producing a high-quality drivable space output for automated driving.