Getty
It’s not just state and federal regulators’ safety concerns that are preventing automotive manufacturers from producing self-driving cars for the masses. First they have to figure out how to make them more energy efficient.
Opening the cargo area of a typical self-driving test vehicle usually reveals a trunk full of computers and wires needed to process petabytes of sensor data in real-time. That doesn’t leave a lot of room for luggage, groceries, or anything else you usually transport in a car, not to mention the huge amounts of energy these systems suck from the batteries as they process all this information.
That’s why BMW I Ventures and Toyota AI Ventures has invested in Recogni, a San Jose, Calif.-based startup that is developing an artificial intelligence platform optimized for autonomous vehicles that can process information quickly while consuming very little energy.
Recogni is developing a computer vision-based inferencing system that uses a diverse set of image sensors that can identify significantly smaller objects at a much larger distance compared to competitors. Using a camera-based approach to developing autonomous driving platforms isn’t new, but according to Ashwini Choudhary, cofounder and chief business officer, what separates his product from other systems is that it’s also simultaneously developing its technology to be ultra low-energy consuming.
“As cars will be doing the actual driving, they will need vast amounts of computation power at very high efficiency. Even if you get clever with the slew of current accelerator architectures, the vehicle is still consuming tens of kilowatts of energy, which is a nonstarter already,” wrote Choudhary in the company’s blog. “Recogni technology will change that entirely and get the work done for the entire car in less than 100w, while processing the tsunami of data in realtime.”
It’s this combination of high processing at low power levels that has caught the attention of Silicon Valley and automotive venture capital groups.
“While scoping the market, we realized that most of the neural network accelerator technologies are either optimized for performance or power – none are optimized for both,” said Ashok Krishnamurthi, Managing Partner at GreatPoint Ventures, which led the financing round in which Faurecia, Fluxunit, and DNS Capital also participated.
Recogni initially plans to use their technology to improve perception processing for Level 2 self-driving technologies that many automotive manufacturers already use advanced driver assistance systems. Its Vision Cognition Processor will efficiently solve the “endpoint inferencing problem” with AVs and possibly accelerate the roll-out of level 3, 4, and 5 self-driving cars.
That said, the company believes robotic taxis won’t be economically and technically feasible until 2024, with full self-driving vehicles for personal rolling out a few years later.
In the mean time, Recogni will use the $25 million it raised from the Series A financing to grow its engineering team and continue to develop its inferencing system.
