In CleanTechnica‘s exclusive, in-depth interview with Peter Mertens, former board member of Audi, Volkswagen Group, Volvo, & Jaguar Land Rover and former R&D head of Audi (published yesterday), Dr. Mertens briefly discussed a few startups he’s now on the board of. One of those was Recogni, and no offense to the others (which I’ll come back to in another article), but Recogni seemed to be the one that got him most lit up.
As Mertens stated at one point, his initial thought when being introduced to the company was, “Okay, wow, these guys are completely crazy! I mean, what they [are trying] to achieve is almost impossible. But if they could achieve it, [it’s] really gonna be a breakthrough — in terms of vision processing, enabling autonomous drive in a way that no one has ever thought it was possible. And then I met the guys and I said, ‘they can do it,’ and guess what — I mean, with proof of concept right now — they will deliver.”
That is all he said about Recogni, but it was perhaps the most excited he was in the whole one-hour interview (which I highly recommend watching), and his statements were quite grand, unlike many other more cautious or measured statements on other topics.
With such excitement, I wanted to look into this. First, though, a couple of notes on why Recogni now has a leg up without even looking at any of its tech. As noted previously by Alex Voigt and basically confirmed in his interview with Dr. Mertens, following recovery from a health matter, Dr. Mertens was basically being offered a CEO position somewhere under the Volkswagen Group umbrella. He declined returning to corporate life, strictly for personal reasons it seems (to not overwork himself in such an environment again), but his opinion on these matters is surely still valued very highly within the many walls of Volkswagen Group. If he recommends a startup like Recogni to the automotive giant at some point for acquisition or investment, without even glancing at anything else, they would take the recommendation seriously and look with an open mind (or even eager mind) into the suggestion. Naturally, I skipped over the part of simply opening the door to Volkswagen Group and several other automotive giants — which many startups struggle to do — because this would go beyond opening a door; it would be a recommendation from someone who probably would have been the final decision maker on the topic if he stayed at the company.
With that corporate consideration out of the way, let’s see what we can find on the tech itself. We can start with a few more comments (biased, naturally) that the company highlights on its website from lead investors, including from the AI investment arm of the other largest automaker on the planet:
“Autonomous systems are becoming smarter, driven by more powerful edge processing. The next opportunity is to achieve this higher machine intelligence at much lower power. We are excited by Recogni’s inference architecture for high-performance, low-power AI computing at the edge, and look forward to working with the team to build a world of safe and efficient autonomous systems.” — Jim Adler, Founding Managing Director of Toyota AI Ventures
“The ability to process sensor data on the edge efficiently and in real-time is essential in the development of autonomous vehicles. We believe that Recogni has the right approach and an experienced team to help solve these critical issues as the automotive industry continues on its path towards semi-autonomous and fully autonomous vehicles.” — Marcus Behrendt, BMW i Ventures
“We truly believe in the sensor fusion based on camera, radar, and lidar, but computational requirements for those algorithms remains one of the critical bottlenecks in autonomous driving today. Recogni solves this problem with a unique and disruptive approach — we are proud to back this team of world-class IC and system developers, as well as automotive AI experts.” — Sebastian Stamm, Fluxunit — Osram Ventures
As you can glean just from these company-highlighted comments of praise, this tech is based on improving the efficiency of computational processing of sensor data (“edge processing”). Recogni’s claim is that it processes an enormous amount of data using very little power. Recogni CEO RK Anand summarizes the challenge and solution in his own words:
“The issues with the Level 2+, 3, 4 and 5 autonomy ecosystem range from capturing/generating training data to inferring in real-time. These vehicles need datacenter class performance while consuming minuscule amounts of power. Leveraging our background in machine learning, computer vision, silicon, and system design, we are engineering a fundamentally new system that benefits the auto industry with very high efficiency at the lowest power consumption.”
In three quick notes, the company explains in more detail what it is processing, how it is unique, and where in the AI-vehicle system it is operating:
- “It’s the only multi-ocular camera system architecture purpose-built for object recognition that extracts passive stereoscopic depth at the pixel level.”
- “Recogni achieves greater processing efficiency & speed by storing weights (parameters) of the object library on-chip, where the computational analysis is performed.”
- “Recogni’s module is pipelined and operates at greater than 8Mpixel images at 60 frames per second, where it is able to recognize (detect, segment & classify) objects, fuse depth-sensor information into the objects, and provide the intelligence to the central system within a few milliseconds.”
Tesla’s software and autonomous vehicle hardware leadership is often mentioned superficially, with basic language any mere mortal like me would understand, but when you dive deeper, a few things have become very clear:
- Tesla excels in vehicle efficiency — because its leadership understands how critical this is for both electric vehicle range and for achieving Full Self Driving. You need to preserve as much energy for autonomous driving processing as possible.
- Tesla’s whole vehicle is built around the computer hardware and software inside. It is the only major automaker that basically builds computers and puts a vehicle architecture around them rather tagging small computers onto an old-school automobile design here or there.
- Tesla is extremely vertically integrated when it comes to all of these matters, whereas conventional automakers outsource almost all of their computing needs (hardware and software). Volkswagen is trying to change in this regard, but seems to be having problems. Inside of Tesla — a core part of Tesla — is an autonomous driving startup like no other automaker has.
- Tesla is constantly, obsessively collecting as much data as it can from vehicles on the road in order to improve its self-driving systems (hardware and software).
The next three (last three) quotes from Recogni further highlight the importance of those above points:
- “True driverless vehicles must analyze the environment, recognize objects at a distance, and make a decision in less than 50 milliseconds for urban driving and less than 30 milliseconds for highway driving.”
- “Latency constraints require all image processing to be done within the car’s systems.”
- “Cars have limited energy and power they can devote to the computational tasks without affecting range.”
George Hotz, another player in this space with his own AV-tech startup, made similar points to me in 2017 when explaining Tesla’s leadership in the field, laughing about what other automakers were doing (some harsh criticism in words and in tone), and sort of pitching his own efforts. (If you watch the video below, note that he was supposed to be at the Paris conference in person but missed his flight and was thus broadcast live onto the projector screen — I was not filming my computer screen.)
Tesla fanboy, Tesla fanboy … I thought this was about Recogni? Yes, this is still about Recogni, and trying to put it in context, which makes it important to explain what Tesla is doing that’s so different from other automakers. Don’t believe me? Then read some thoughts from Chief Business Officer and cofounder of Recogni, Ashwini Choudhary, who had a section of his latest article in Forbes titled “The Tesla Way.”
“Tesla has taken a multipronged approach to mete out its assault on the auto industry. First, the company built an electric-powered car that is fun to drive, and then it gave the driver a user experience that almost feels like a smartphone.
“Second, Tesla is essentially a software company. While the old auto industry is playing catch-up to the new user interface, Tesla is taking the fight to them with autonomous vehicles. The company is miles ahead of the rest of the industry (no pun intended), both from a technology and a marketing perspective. It is ironic to look at this giant industry from this perspective, but it’s necessary.
“An electronic, technology-centric approach to making a car is the disruption the auto industry needs to survive. Tesla is a technology company that makes cars. The company is driving innovation at an exponential rate with periodic remote software updates that change the personality of these cars. Other companies are manufacturing firms that make cars but don’t understand technology as Tesla does.”
Perhaps not the way one would invite themselves into the home of someone they wanted to marry, but Choudhary’s point for automakers is that Tesla is solving the automotive problem of the day for itself and automakers better find someone to help them solve that problem ASAP as well.
While it must be 100% clear at this point that you can’t add bolt-on autonomous-driving solutions to a vehicle not designed at its roots for autonomous driving, there is still debate about whether vertical integration or horizontal integration is best for this evolution. Alex Voigt has argued very strongly on CleanTechnica that vertical integration is the way to go. Recogni’s case is the opposite:
“[E]lectronics technology is not [automakers’] core competency, and just throwing a large team at the issue will not solve the problem. These companies need to make a bet on the right partnerships. Either the companies go completely vertical, similar to Tesla, and develop the entire technology stack themselves — which will be difficult for the incumbents given their manufacturing DNA — or they go horizontal and source ‘best of breed’ technology from various entities and integrate them intelligently. The second approach will enable automotive OEMs to leapfrog Tesla altogether.”
The ability to beat Tesla is a bold promise, but it’s one that Choudhary repeated in his late-January article in Forbes: “What I saw recently at CES 2020 gives me hope. I met several CTOs of automotive OEMs and tier-one parts suppliers who have the vision and want to beat Tesla by going horizontal.”
I’m excited to see how this plays out. I’m a fan of creative and innovation-inducing competition, not monopolies. Whether horizontal integration can beat vertical integration is yet to be proven, but even if it can, it is going to depend on a stellar team leading the integration and in primary control of vehicle development. Let’s see who can put together the best teams and the best tech.
To close, Recogni has just two press releases in its docket, an unstealthing press release on July 31, 2019, and an announcement in early November 2019 of Peter Mertens (see top of article) joining the board. However, I expect we’ll be hearing much more about the company in the coming years. I certainly look forward to reaching out for more insight into what it’s offering.