in

How May Mobility’s autonomous shuttle ambitions backfired

How May Mobility’s autonomous shuttle ambitions backfired
Ann Arbor startup May Mobility’s autonomous shuttle.

By all appearances, May Mobility was a scrappy success story. The autonomous transportation startup made its debut at Y Combinator’s demo day in 2017, with a team that had been working on driverless tech since the third U.S. Defense Advanced Research Projects Agency (DARPA) Grand Challenge in 2017. Within the span of a few years, May had a roster of paying customers in Michigan, Ohio, and Rhode Island as it raised tens of millions in venture capital from investors including Toyota and BMW.

But on the inside looking out, it was a different story. May engineers struggled to maintain and upgrade the company’s vehicle platform, at one point spending months attempting to install an air conditioning system in the depths of summer. The leadership’s ambition often outstretched May’s ability to deliver, which upset vendors, some of whom went unpaid for stretches. And not a single one of the company’s commercial routes approached full autonomy.

Conversations with former May employees reveal a startup struggling to stand out in an industry dominated by incumbents like Waymo, Uber, Aurora, Cruise, and Amazon’s Zoox. As one source put it, May’s intent might not have been malicious — executives at the top were convinced it would succeed. But overeagerness and inexperience led to missteps that soured municipal relationships.

Auspicious beginnings

May Mobility was cofounded by Edwin Olson, who previously acted as lead investigator on Ford’s driverless car program and co-directed autonomous driving projects at the Toyota Research Institute, Toyota’s mobility- and robotics-focused R&D division. An associate professor of computer science, Olson assumed the role of CEO at May after taking leave from the University of Michigan, joining GM Ventures veteran Alisyn Malek, May’s COO.

Malek, like Olson, came from the autonomous car industry. At GM Ventures, she oversaw the relationship with then-startup Cruise, which is now developing GM’s driverless car platform. Steve Vozar, May’s other cofounder and chief technical officer, was also a member of Ford’s driverless vehicle program and directed the University of Michigan’s APRIL robotics laboratory.

From the start, May’s aim was to deploy implementations of autonomous driving technology in the real world, with an emphasis on what’s possible today as opposed to 5 or 10 years into the future. In alignment with that goal, the company targeted customers with fixed-route transportation needs in geofenced, easily mappable business districts, campuses, and closed residential communities.

“We’ve seen the Ford, Toyota, and GM experiences up close,” Olson told TechCrunch in an August 2017 interview. “Like the other big companies in this space … they’re all after the transportation market as a whole. This is the $4 trillion problem of transportation on demand, think autonomous Uber. [W]e think that by getting out into the real world, we can not only build a successful business, but we can turn on a flow of data and operational know-how that will help us move and improve our systems faster than the [original equipment manufacturers] which are in R&D mode.”

May chose not to design vehicles itself but instead focused on software, low-level electromechanical systems, and sensors while sourcing and building upon third-party hardware. Its “secret sauce” was ostensibly a set of algorithms the company called Multi-Policy Decision Making, work on which began in Olson’s lab. Instead of training the system by feeding it data representing many different scenarios and teaching it to react to those scenarios, May’s software was designed to understand situations and predict what’s happening on an agent-by-agent basis.

To further differentiate itself, May launched a full fleet operation service covering everything from daily operation to maintenance and cleaning. It also pledged to work with customers to integrate its autonomous stack with their existing fleets.

“May Mobility, like Waymo, Cruise, and Uber, [is] building full-featured technology stacks designed for the full range of autonomous driving scenarios — even if our first markets will be more structured,” Olson told Xconomy in a March 2018 interview. “Google, Uber, and Cruise have all invested massively in their teams, but workable unit economics — which require no safety driver and affordable vehicles — still appear to be years away. We expect to beat all three of them to market with a successful product.”

Hardware challenges

May’s nimble development cycle enabled it to sign on its first clients in 2018, only a few months after its founding. But according to sources, it also placed a strain on the company’s engineering team, which was tasked with retrofitting a vehicle platform that was not designed with May’s deployments in mind.

May used the drivetrain and chassis of Polaris’ six-seat, shuttle-like GEM, which has a top speed of 25 miles per hour, allowing it to be used on public roads without approval. But in cities like Columbus, Ohio, the modified GEM struggled to handle inclement weather conditions. May had to cancel operations on some days during the winter because the low temperatures prevented the GEM’s batteries from starting.

Even in better conditions, May’s fleet availability became a problem as multiple shuttles broke down at once. The pilot in Providence, Rhode Island, which had a dozen vehicles on-site, saw only half of those vehicles transport passengers at any one time while the rest were being maintenanced.

And despite its Level 4 autonomy target, May struggled to achieve self-driving in the cities where the GEMs were deployed. That’s in spite of the fact the shuttles were guided by additional, out-of-vehicle sensors mounted on traffic lights that supplied info like street light color.

Sources say this was a result of difficult driving environments and the GEM’s shortcomings.

After 500 hours of testing at the Quonset Business Park in North Kingstown, Rhode Island, humans had to override the GEM’s systems for left-hand turns with traffic, right turns on red, rain, wind, pedestrians standing at crosswalks, construction work zones, certain four-way intersections, speed bumps, potholes, and aggressive drivers. Foliage also caused complications because the preprogrammed route was mapped when trees were bare. Once the leaves and other foliage emerged, it disrupted the camera, radar, and lidar sensors that helped the shuttles to navigate.

One source described May’s autonomy as “all over the map.” While some days saw autonomy rates hit 80% to 90%, performance often dipped “well under” that mark.

May director of marketing Laurie Remias told VentureBeat the company takes into consideration all conditions the GEMs encounter when measuring autonomy rate. The numbers are “highly seasonal” and depend on the time of day, as well as on the real-time traffic conditions. “If it’s snowing, [our autonomy rate is] 0% because we do not operate our vehicles autonomously in poor weather conditions,” she said via email. “It’s very important when talking autonomy rates to discuss them in full context.”

The Providence shuttles also had to contend with another complicating factor: high-speed traffic. In some areas where they operated, drivers traveled up to 50 miles per hour, far exceeding the maximum speed limit of the GEMs — which lacked an airbag.

Beyond airbags and air conditioning, the latter of which wasn’t installed in the GEM shuttles until August 2019 owing to the need for dedicated batteries, the GEMs lacked a heating system. Engineers were forced to add a diesel-powered heater to the platform, which meant they wouldn’t be fully electric — one of their original selling points.

“The industry is really pushing this at a level that is inappropriate and misguided,” Julia Gold, Rhode Island Department of Transportation’s chief of sustainability and innovation, told EcoRI News in an interview last year. “[May] has not been able to show it’s made fully autonomous vehicles safer than human drivers … We are pushing them to do better.”

Operations missteps

May’s stumbles didn’t end with engineering. Operational challenges resulted in lower-than-anticipated ridership, with most rides — which were free during pilots — falling short of break-even. One source pegs each deployment’s losses at over $1 million a year.

In Columbus, where May’s GEM shuttles traveled several miles through the Scioto Mile district, the city spent an estimated $30 per passenger on one 1.2-mile loop, with May paying about $90 per passenger. As a result of this and the route’s poor autonomy rate, in August 2019, the city sent a 30-day notice letter to May threatening cancellation and legal action. Columbus is now pursuing action to recover a portion of contract money for what it alleges were service-level agreement violations.

A source described May’s Columbus deployment as more a “showpiece” than a functional commuter route, with “tourist attractions” like the Smart Columbus Experience Center, a showroom of the city’s smart city efforts. While the shuttles were available nearly every day during daylight hours, there “wasn’t a high uptake of people using the service,” the source said.

Tellingly, Columbus opted to contract a May competitor, EasyMile, for its second autonomous route.

In Providence, where routes had to be suspended for several hours on hot days because of the air conditioning issue, one shuttle was involved in a fender bender with another shuttle. There weren’t any injuries, but sensors were damaged.

And when the city and Columbus asked May to bring its vehicles into compliance with the Americans with Disabilities Act (ADA), the company dragged its feet, taking weeks to roll out a second-generation shuttle with a wheelchair restraint system and a retracting ramp. In Providence, it also refused rides to children who didn’t bring along a booster or safety seat.

Ridership wasn’t any higher in Providence than in Columbus, even as May attempted to undercut public transit. The average ride across more than six miles in Providence — the entire route — was less than one passenger, not including the fleet attendant. Between May and August, about 120 riders used the service daily, with daily ridership reaching a peak of 260 in June 2019.

Grand Rapids, Michigan, the site of another May pilot, was the one bright spot. As of March, ridership on the 3.2-mile, 20-plus-stop route remained steady nine months into the program at 500 rides a day, passing over the 50,000 total passenger mark before the pandemic forced a pause in operations.

May announced that it reached 100,000 total rides in August 2019, but it hasn’t shared an update since then.

Failed fundraising

May’s deployments were paid for with a mix of private and public capital. For the Columbus route, contributions came from the Smart Circuit program backed by Smart Columbus, a public-private partnership between a corporate group and city government. (In 2016, the U.S. Department of Transportation awarded Columbus the designation of Smart City, furnishing it with a $40 million infrastructure fund.) Rhode Island spent at least $800,000, some of which came from settlement money the state received from the Volkswagen emissions scandal and the Federal Highway Administration.

But the model wasn’t sufficiently lucrative to sustain May over the long term, which spurred the company to seek backing from strategic partners. When this failed to materialize — the company’s series A came to $22 million, far short of its $100 million moonshot — it took drastic steps to cushion the blow.

As May’s C-Suite pursued a series B and considered closing an extension to the series A round, they made decisions to cut back on payments to vendors to stretch the burn rate as much as possible. In what a source called “the perfect storm” of funding pressures, May delayed payments to partners “wherever possible.” The mandate from the top was to look for opportunities to slow payments, with the idea being to hold onto payments longer than invoices were due.

In a statement, Remias denied the company failed to pay vendors, claiming it “worked closely” with a few of its suppliers to set up payment plans. She added that as of today, all of these payments have been “entirely satisfied.”

Around this same time, May began putting together a projection for Toyota on how it might expand service from four cities — Providence, Detroit, Columbus, and Grand Rapids — to more than 80 by 2022. Revised pitches referenced deployments in Tokyo, Toyota’s home city, where May envisioned providing transportation for the now-postponed Olympic Games.

“It was anyone’s guess how May would accomplish that,” one source said.

In an implicit vote of no confidence, between December 2019 and January, May lost three out of four of its C-Suite members — the COO, CTO, and CCO — as well as the heads of reliability and sales and marketing. The company’s head of product left last summer.

“We have had a few employees leave the company over the last year to pursue new opportunities,” Remias said in response to questions about the executive departures. “That said, as we position ourselves for the future, we’ve also been growing our executive team and have recently hired a VP of software, VP of people operations, chief product officer, chief engineer, and VP chief of staff.”

Good intentions

After its pilots in Columbus, Rhode Island, and Grand Rapids came to a close and planned deployments in Florida and Texas fell through, May redoubled its efforts in Detroit with Bedrock, the real estate firm associated with Quicken Loans chairman Dan Gilbert. It was recently announced that GHSP, a company developing an ultraviolet-C treatment that automatically disinfects the air and high-touch surfaces within vehicles, would receive $80,000 from the Michigan Economic Development Corp.’s PlanetM program to install the treatment in May’s shuttles.

Had May’s rollouts gone smoother, the company intended to explore connecting routes to public transit as it expanded its per-city fleet size to 25. In communities like Detroit and Columbus, the company hoped to replace diesel-powered buses with its mostly electric alternative, targeting the half of U.S. trips that are three miles or less.

“Our ideal partners are people that have first mile, last-mile challenges, who are trying to help individuals get from things like transit stops or parking structures to their end destination,” Malek told Wired in December 2018. “What this gives us, essentially, is a captive audience, and a known road network, and nodes that people need to move between.”

May’s road post-deployment deployment has been bumpier than most, but some of its rivals haven’t fared much better. In February, the National Highway Traffic Safety Administration (NHTSA) partially suspended U.S. operations of France’s EasyMile after a passenger in Columbus was injured while riding in one of the company’s driverless shuttles. And in 2018, the NHTSA suspended a separate Transdev program in Florida that sought to replace school buses with EasyMile’s vehicles.

The Department of Transportation two years ago published a prescient report on the autonomous shuttle sector, highlighting the limited vehicle autonomy, procurement challenges, and regulatory unpredictability shuttle startups have yet to address. “The market is small, and many companies in this space have little experience designing and validating systems and producing vehicles, compared to traditional automakers,” the report’s authors wrote. “Low-speed automated shuttles may not be suitable for all environments and services.”

The industry isn’t without apparent success stories, like Optimus Ride. But May’s setbacks illustrate the nascent technology’s limited applicability, particularly when stretched beyond its capabilities.

Source: venturebeat.com

What do you think?

486 points
Upvote Downvote

Leave a Reply

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.

Going Driverless: A look into the future of the trucking industry

Going Driverless: A look into the future of the trucking industry

Horiba Mira offers simulation platform to autonomous car developers

Horiba Mira offers simulation platform to autonomous car developers