Is driverless car on same route as 3D TV? – Part 1
Article By : Junko Yoshida
From start-ups to leading tech companies, opportunities abound for everyone to pile onto the self-driving bandwagon. But there are endless questions about driverless cars.
Let’s face it. The drumbeat of hype for self-driving cars isn’t slowing down. It’s only getting louder.
We hear less talk about a more reasonable, incremental approach for autonomous cars. Automakers and technology suppliers grow more aggressive and accelerate their plans on an almost daily basis.
From leading chip vendors to little known start-ups, it appears that nobody wants to miss the seemingly once-in-a-life-time opportunity to pile onto the self-driven bandwagon.
Carpe diem. I get that.
Earlier this week, I talked to Laszlo Kishonti, founder and CEO of AdasWorks, a Budapest, Hungary-based start-up. The company, according to Kishonti, develops “a full stack of self-driving car software” operable on any processor–GPU, FPGA or embedded. “We are processor-agnostic,” he said.
Kishonti was visiting California, with plans to open an office in Silicon Valley by the end of this summer. He said that while the company’s R&D team remains in Budapest, a Bay Area-based office will be crucial for business development, support and testing AdasWorks’ software for self-driving cars running on real streets in the United States.
“We will have at least two or three cars that work in a full self-driving mode” for such road tests, said Kishonti.
The company raised ₹16.89 crore ($2.5 million) last year in seed funding from investors including Day One Capital Fund Management, Inventure Oy and Robert Bosch Venture Capital. It has increased its funding to ₹67.57 crore ($10 million) in recent months, as part of an extension to the Series A funding from the same investors, according to the CEO. “The proper Series B funding is scheduled later this year,” he said.
As far as AdasWorks is concerned, the promise of self-driving cars is raising real money, allowing the company to come to Silicon Valley and exploit its business prospects.
In contrast, the company that sits almost at an opposite point on the autonomous car hype cycle is Nvidia.
Nvidia is where start-ups like AdasWorks are aiming to be.
[As a side note, AdasWorks’ software for Intelligent Surround View System with 4 cameras is being used in 100 self-driving Volvo XC90 vehicles–based on Nvidia’s Drive PX platform—scheduled for testing in Gothenburg in 2017 in the Drive Me project.]
Nvidia has jumped the gun on the fully autonomous car debate by pitching “a centralised supercomputer inside the vehicle.”
To its credit, Nvidia was the first to educate the world on “deep learning” in the context of autonomous cars. The company has practically rebranded itself as a supplier of “the world’s most advanced autonomous car platform that changes the driving experience.”
Nvidia has also successfully milked that reputation. According to Nvidia, the company currently has “over 80 OEMs, Tier 1s, and research institutions to develop ADAS and autonomous solutions using our DRIVE PX 2 supercomputer.”
When we asked Danny Shapiro, Nvidia’s senior director of automotive, about the automotive industry’s accelerated schedule for fully autonomous cars, he said, “When the autonomous car gets on Time magazine’s cover, or gets featured in Esquire, no CEO at automakers wants to be left behind.”
The analyst community often describes Nvidia’s Drive PX 2 as an R&D platform for fully autonomous cars. Shapiro, however, begs to differ.
“When it comes to fully autonomous cars, no automakers are expecting to get a black box,” he explained. Tier ones and OEMs, perhaps, want to experiment, tinker and design their own architecture or software.
Describing Drive PX 2 as a “scalable and flexible platform,” Shapiro stressed that Nvidia’s customers—automakers, tier-one suppliers and start-ups developing self-driving cars—can test their ideas and scale the system.
The processing power of Drive PX 2 “can scale from a single mobile processor running at 10 Watts, to a four processor (two mobile processors and two of NVIDIA's highest end GPUs) delivering 24 trillion deep learning operations per second,” the company said. “The fully configured system could be up to 250W, including options for cooling with air or liquid.” In short, “It all depends on what the OEM wants,” said Shapiro.
The hitch remains, however, that the fully autonomous, driverless car is damn hard to develop.
As for real hurdles waiting for self-driving cars, I’ve come across several astute observations from industry observers this week alone.