As Zoox, the secretive robotaxi developer recently acquired by Amazon, gets ready to unveil its futuristic fleet vehicle, its former CEO who dreamed up the company is re-emerging with a new startup that’s designing AI-enabled software he hopes will allow cars to “teach themselves” to drive.
Early-stage HYPR, created by Zoox cofounder Tim Kentley Klay, says it’s using reinforcement learning, a branch of machine learning that utilizes a reward-based approach, to train driving algorithms dynamically–ideally with no need for direct human instruction or supervision. The Alameda, California-based startup has raised a $10 million seed round and begun testing its approach with a modified Daimler Smart Car. Backers include R7 Ventures and Australian billionaire Andrew Forrest.
Kentley Klay “kicked in a modest amount,” he says without elaborating.
“HYPR will be an ahead-of-the-curve robotics company,” Kentley Klay tells Forbes. “We are focused on building the learning software stack and infrastructure. When we have that, we will share how we are going to exploit it.”
As a designer, commercial artist and entrepreneur hailing from Melbourne, Australia, rather than an auto or AI engineer, Kentley Klay was an improbable entrant into the crowded race to develop self-driving cars when he cofounded Zoox with Stanford University computer scientist Jesse Levinson in 2014. The pair raised nearly $1 billion for a pure robotaxi company serving big cities, starting with San Francisco, pursuing Kentley Klay’s sci-fi vision for ferrying passengers in purpose-built, driverless electric vehicles that maximize rider comfort and convenience–and with no steering wheels, brake and accelerator pedals.
Disagreements over strategy led to Kentley Klay’s unexpected firing in late 2018. The Foster City, California-based company was acquired by Amazon this year for an estimated $1.3 billion. Zoox cofounder Levinson remains chief technology officer.
While expectations for the potential of driverless technology remain high, solving its technical challenges is taking longer than many expected. The Covid-19 pandemic has further complicated things this year. Industry leaders including Alphabet Inc.’s Waymo, General Motors-backed Cruise, Motional and other programs are readying or already operating limited public ride services to demonstrate the technology but wide-scale availability of on-demand robotaxis is unlikely for the next few years.
(For more on the company, see The Wild Ride Of Zoox from May 9, 2018.)
For its part, Zoox is to finally unveil an in-house-designed, bi-directional robotaxi next month that will be built at a production facility the company has set up in Fremont, California. New CEO Aicha Evans hasn’t said when commercial operations will start.
With HYPR, Kentley Klay wants to try a new approach to perfecting the AI behind the wheel, ideally one that’s faster than through techniques and testing methods used by established AV companies.
“I believe in end-to-end learning systems that are cloud connected for distributed online learning, in parallel. I believe in the near future, this is how all robotic systems will be formulated. And if that is true, then at scale, we are talking about cloud centric neural networks coalescing robotic assets deployed in the trillions,” he says.
“Stay tuned: it is going to be both unexpected and unprecedented.”