I met Stefan Seltz-Axmacher for the first time in November 2015 at the Florida Automated Vehicles Summit. Not long after, we met at the Blue Danube coffee shop in Alameda, CA so he could tell me about his vision for Starsky Robotics. When he energetically described his remote-driving-for-trucks approach, I was skeptical. “Remote driving is hard,” I said. “The military has struggled with this for years. Its harder than it looks.” On the technical side, latency for secure communications is challenging. On the operational side, re-creating enough on-road reality (situational awareness) for a remote driver is difficult when going for the high levels of safety needed. Seltz-Axmacher remained bullish on the approach and at that time went on to found Starsky Robotics as one of the earliest truck AV startups, later closing a $16.5M Series A funding round in March 2018, and then hauling freight while developing both remote and automated driving ability. Initially, Starsky’s concept was all about remote driving for first/last mile. They later expanded their offering to include fully automated highway driving on limited freight corridors.
Now, Starsky has become the first casualty within a crowded truck automation space, and Seltz-Axmacher has provided us with an intriguing post-mortem in a recent Medium post. Most of the media coverage I’ve seen has acted as echo chambers for Seltz-Axmacher’s perspective. Here I offer a counterpoint based on my longtime involvement in truck automation plus discussions with many others in the truck Automated Driving Systems (ADS) startup space, many of them irate at what they see as unfounded assertions made in the original post. My sources tell me that because Seltz-Axmacher hasn’t experienced their technology nor been briefed on their technical/safety approach, he has no basis to make sweeping claims about the entire industry.
Starsky was founded around the idea of remote driving, i.e. moment-by-moment operation of the vehicle by a remote operator. Their original premise to relocate the driver from cab to office was of interest at the time. Due to the pressures on truck fleets from the driver shortage, hiring and retention would be easier if they could guarantee their drivers would be home at night rather than on the road.
Remote driving was intended to bridge between the reality at that time and future fully automated driving. However, as technical development began to enable truly driverless trucking, the potential to replace, rather than re-locate, the driver created significant excitement. That is what spurred the founding of more startups during 2017-2019. Starsky also went in that direction to some degree, but it was a crowded space. Investors saw quick progress in full automation which took the shine off remote driving. The current set of active companies in truck ADS are shown in the image below.
Can AI Get The Job Done?
One of Seltz-Axmacher’s key premises is that the development of automated driving systems (ADS) is moving slower than was originally expected a few years ago, asserting that the performance gains from traditional artificial intelligence (AI) based on supervised learning are starting to stall out. As his argument goes, investors have become disillusioned as a result. He predicts a long road to reach adequate capability for safe on-road driving, a road he says is likely too long for other startups to survive (noting that he would “love to be wrong” about this last point).
There’s no question investors did moderate their enthusiasm and want more substance last year. 2019 was a re-set year in expectations for automated driving generally. It wasn’t just the technology, there were questions within OEM boardrooms regarding the business case. Several passenger car OEMs pulled back from redefining themselves as mobility service providers. Daimler CEO Ola Kaellenius spoke of the need to “rightsize” spending on robotaxis “due to increased costs and regulatory hurdles,” noting that “ride-hailing return-on-investment is unclear.” And even though Tesla continues to push the limits, traditional OEMs moderated their pace of moving up-level in automation capability for vehicles marketed to the public. As Hakan Samuelsson, Volvo Cars CEO, succinctly put it, “Automated driving is a bit more challenging technically than we originally thought.” GM is an exception, staying the course on robo-taxi but Cruise Automation did not deploy robotaxi in 2019 as planned and they have not provided a new timetable.
Nevertheless, in 2019 we passed the mind-blowing milestone of truly driverless vehicles taking regular people on rides via Waymo’s ride-hailing service in Chandler, AZ. Remotely-driven driverless vehicles from Einride and Nuro began moving goods on surface streets, traveling at low speeds such that data delays from communications latency are much less of a challenge. (Note that, being driverless, these are the only ride hailing and delivery services that are COVID-19 “safe” on the planet.)
The upshot of the “2019 reset” is a broadly held view that driverless trucks are the best bet, as also noted by Kaellenius. Originally, commercial development of truck ADS began as truck startups going it alone, and now OEMs such as Daimler, Traton, and Volvo Group, among others, have ramped up ADS programs and funding.
Some respected AI experts agree that the limitations of supervised machine learning are creating challenges. They are advocating advanced techniques such as unsupervised machine learning. The dynamics Seltz-Axmacher refer to are present, but he takes the trend much too far, saying “The space was too overwhelmed with the unmet promise of AI to focus on a practical solution. As those breakthroughs failed to appear, the downpour of investor interest became a drizzle.” As one echo-chamber article noted, SSA’s post “painted an ominous sign for the autonomous tractor-trailer industry”.
Not so, say others in the truck ADS space who I spoke with. Maybe those breakthroughs “failed to appear” for Starsky, but a spokesperson for TuSimple said “We believe our solution is advancing well and the AI approach we’ve taken is delivering well on its promise.”
Plus.ai agrees that Starksy’s experience is not representative of the entire truck ADS space. Shawn Kerrigan, COO and co-founder, said “The capital markets are becoming more selective when investing in autonomous trucking companies, which is wise considering that several early market entrants over-promised and under-delivered. While the number of autonomous trucking companies is now smaller as the market matures, the amount and pace of investment going into that smaller group of companies is increasing. Most importantly, the broader ecosystem’s commitment to autonomous trucks remains unchanged —the world’s largest OEMs, Tier 1 suppliers, and regulators continue to invest in and prepare for the launch of autonomous trucks in several years.”
Consensus: Lack of Investment in Starsky Was About Starsky
While Starsky was seeking to raise funds last summer, they were competing with Embark, which raised an oversubscribed Series C at $70M for self-driving trucks. Earlier in 2019 Outrider raised $53M in their Series A. In 2018, IKE raised $52M and Kodiak Robotics raised $40M Series A’s, respectively. More recently, we see that Waymo just announced a $2B raise, and Pony.ai raised $400M last month. Pony.ai is focused on robo-taxis, whereas Waymo is involved in both robo-taxi’s and robo-trucks. The “failure of AI” narrative addressing investor sentiment used by Seltz-Axmacher applies to both, and based on these investment levels doesn’t stand up to scrutiny. Plus, Seltz-Axmacher’s comments do not even mention the massive internal investments made recently by Daimler, Traton, and Volvo Group in the truck OEM space.
Seltz-Axmacher goes further, saying “Unfortunately, when investors cool on a space, they generally cool on the entire space.” This is not the view at all for the majority of ADS trucking startups. The TuSimple spokesperson noted “We don’t believe investors have cooled on this space. On the contrary, we see significant investor interest.” As another ADS company put it, “several other companies successfully raised money during and since Starsky’s failed round, which suggests the problem is less with waning enthusiasm for the industry, and more with Starsky specifically.” Yet another noted “we’ve seen from Amazon to Uber and everywhere in between, VC investors are willing to give startups a lot of money and many years to reach revenue targets/profitability if they think the company is sufficiently focused on a business idea that has significant value in the long term. Everyone is aware that the last 10-20% is by far the most difficult.” The clear consensus across the truck ADS industry is that the investment climate has been robust, notwithstanding the long term (and unknown) economic effects of COVID-19.
“No One Really Likes Safety”
Maybe just to be provocative, Seltz-Axmacher titled an entire section with “No One Really Likes Safety, they like Features.” Referring to Starsky’s heavy emphasis on good safety practices, he goes on to say “the problem is that all of that work is invisible. Investors expect founders to lie to them — so how are they to believe that the unmanned run we did actually only had a 1 in a million chance of fatality accident?”
Locomation CEO Çetin Meriçli observed that Seltz-Axmacher’s assertions “regarding VCs being more impressed with AI promises is correct, and partly explains why the other companies were able to raise more money.” A contact at another truck ADS company noted “Stefan is absolutely correct in saying that many investors don’t yet fully understand AV technology and have therefore gravitated towards companies with flashy demos. But we see smarter investors in the space start to emerge that are searching for teams that have the experience and maturity needed to build a compelling safety case.”
The question was raised, “Did Stefan think all investors aren’t interested in safety because his investors weren’t interested in safety?”
I was pleased to “go to the source” and get an investor perspective from Matt Ocko, Managing Partner, DCVC. He said “We are anchor, on-going major investors in Embark. DCVC – and the other top-tier VCs who are investors here – focus on results, not promises. We expect safety to be evident in the DNA of an autonomous driving company, and we see it in Embark’s engineering approach and their on-the-road results. Embark has consistently impressed us with their ability to develop automated driving capability and safety in a deeply intertwined manner – while spending in a disciplined way with plenty of runway. Going forward, we are convinced that autonomous trucking continues to be a good investment due to the huge driver shortage and intense pressures on the supply chain.” I expect investors in the other leading ADS startups would speak similarly.
Seltz-Axmacher had more to say. “Our competitors, on the other hand, invested their engineering efforts in building additional AI features.” Seemingly implying “features over safety,” this statement in particular elicited howls of protest. TuSimple firmly stated “Our system was designed with safety as our first priority, not features. We invested heavily in functional and operational safety from the outset. We follow all relevant standards, and are held to this not only by our own interest, but by the interests of our technology partners. We deeply understand our responsibility to our investors, the industry, and to the driving public. This understanding guides everything we do.”
Seltz-Axmacher is proud that in 2019, a Starsky truck was the first fully-unmanned truck to drive on a live highway. At the time, this gave great discomfort to others in ADS trucking and the trucking community overall. This is broadly viewed by others in the industry as a “publicity stunt” to attract investor interest, which endangered the entire industry while providing no engineering value. I’m sure Seltz-Axmacher has a rebuttal, but this is a broadly held view, as noted here: “A lot of what caused Starsky to run into trouble was unique to Starsky. While it’s true that some investors want to see stunts and an ever-growing list of features, the smarter ones are looking for teams that have the experience and vision to bring this technology to market. And while most would agree that a lot of the basic machine learning required to build an AV is fairly well-understood, what differentiates the great teams is having the experience to take a multi-disciplinary approach to building an AV that considers software and machine learning, hardware, functional safety analysis, testing and validation, and building a safety case.”
Seltz-Axmacher’s post over-generalizes about the tech industry and the investor community. Sweeping claims don’t hold up well in this space. The consensus is that investors didn’t abandon all of truck ADS, they abandoned Starsky. This is the hard truth and there’s no other way to say it. The reaction from the larger community was aptly expressed by one source saying, “It looks like Stefan was trying to throw the industry under the bus to avoid taking responsibility for Starsky’s specific failures.”
Nevertheless, I credit Seltz-Axmacher for charging ahead to develop a new market segment and attempt to launch an early form of driverless trucking. Starsky Robotics had “hustle,” which stems directly from his leadership. Most likely he owns a number of patents essential to supporting unmanned vehicles which may be valuable as Remote Assist takes hold in ADS.
Far from being “an ominous sign,” Starsky’s fate is an anomaly not representative of the overall truck ADS space. Is it possible that all the others are blind and Seltz-Axmacher is the only one who can see clearly? Sure. You can listen to this one voice if you want, but I suggest that you pay attention to the comments I received from seven of the leading truck ADS startups. These companies, their investors, and the truck OEMs know that the road ahead is difficult. They’re moving forward with lots of horsepower.