By Manish Singh, VP of Marketing @ Recogni Inc.
The car industry is facing two major evolutions: autonomous vehicles (AV) and electric vehicles (EV). Compared to traditional auto manufacturers, Tesla, who uniquely develops their own purpose-built self-driving platform for their cars, is ahead of the pack with regards to AV and EV technology. Thus, to be competitive and relevant, companies must integrate an industry-leading solution to adapt to this rapidly evolving market.
When driving, our brains subconsciously process a plethora of visual data in real time – just take a look at the figure of the top to see a depiction of this. So, for AVs to become a reality, they must do the same, and quickly and accurately understand their surroundings under any circumstance. However, to do this, AVs need both immense processing capability and great power efficiency to interpret scenes that the camera systems acquire – specifically, they need 75 tera-operations-per-second (TOPS) of processing capability for every watt of power consumption. This enormous efficiency requirement highlights the largest barrier to realizing full vehicle autonomy: the visual scene-understanding problem.
Current solutions for this are based on repurposing legacy technology; however, retrofitting products built for a different application is not optimal – unique problems need purpose-built solutions. Take the smartphone industry, for example. Mobile application processors, with specific functions for various applications, were designed for smartphones and their many capabilities. The nonoptimal solution to this was the general-purpose processor – however, this did not meet smartphone use cases, applications, and power envelope requirements, so an innovative product in the form of a mobile application processor was needed.
Circling back to the automotive industry, because current solutions that serve traditional OEMs are not purpose-built for visual perception, they do not have the processing capability to enable autonomy without compromising battery life. AVs with these solutions today still cannot prevent deadly situations such as head-on crashes or collisions with Vulnerable Road Users. Although Tesla cannot achieve full autonomy today, because their self-driving platform is far ahead of what conventional OEMs are using, other companies need a novel solution that can meet the efficiency requirement necessary to solve the visual perception problem. With this, auto manufacturers can scale from partial autonomy today to full autonomy in the future, allowing them to stay competitive and profitable in the long run as the market evolves.