It took until 2015 before the first true robo-taxis appeared, in the island nation of Singapore of all places.
It took until 2015 before the first true robo-taxis appeared, in the island nation of Singapore of all places.
( Source: gemeinfrei / Unsplash)

Research and Development The evolution of robo-taxis

| Author / Editor: Seth Lambert / Nicole Kareta

For companies interested in developing autonomous vehicle (AV) platforms, the idea of operating robo-taxis has grown increasingly attractive as the prospect of universal Society of Automotive Engineers (SAE) Level 4 or 5 functionality for cars now seems farther away than it did several years ago.

For ridesharing services such as Uber, Lyft, and Singapore-based Grab, robo-taxis are believed to be a major factor in achieving profitability, which for most of these firms has been elusive so far. When AV platform companies set their sights on SAE Level 4 or 5 automation several years ago, it was believed that achieving this goal by the year 2020 or sometime shortly thereafter was all but inevitable. After all, companies like Tesla and Apple were already testing fully automated SAE Level 4 vehicles in places like California and Arizona.

But as time wore on, it quickly became apparent that progress toward Level 4 or 5 functionality for many companies was much slower and more sporadic than expected. Accidents such as the death of pedestrian Elaine Herzberg in 2018 when she was struck by an Uber ATG AV rocked the industry and caused engineers to rethink delivery timeframes for autonomous features and capabilities.

Reassessment has been the order of the day after research companies like Rand Corporation released reports stating that hundreds of millions or even hundreds of billions of AV test-drive miles would be required before company platforms could be considered “safe” for transporting passengers. “Edge cases” (rare, but still-possible dangerous driving situations) still regularly presented themselves to companies doing AV research. Some platforms have not been able to handle environmental factors like heavy weather or intense traffic in all circumstances. In short, there’s still much work to be done—and many months and years to go—before AV platforms can be deemed “ready” for SAE Level 4 or 5 use on every road in every condition.

In the meantime, even deep-pocketed companies like Google sister company Waymo or General Motors subsidiary Cruise have begun to rethink how much money they want to keep throwing into AV R&D. One way to alleviate costs has been to operate robo-taxi services that can help these companies pay for their continued research. As a double function, such services allow these firms to continue their digital learning while they bring in revenue. For ridesharing companies like Uber, Lyft, and Grab, being able to eventually remove the driver from the vehicle means eliminating these firms’ biggest service cost component. To be sure, new avenues of profitability will open up when this can be achieved. Thus, it’s no surprise that these companies are extremely interested in advancing their AV technological capabilities as quickly as possible.

Robo-taxis for revenue

Developing an AV platform is an expensive proposition. Just some of the technical hurdles are integrating camera, radar, ultrasonic, and possibly LiDAR sensors; processing and fusing data from these sources; calculating navigation and trajectories of moving objects and other cars external to the vehicle being driven; and consequently determining steering, acceleration, and braking values to apply during automated driving. All of the computation that takes place has to be processed at the moment input occurs—that is, in “real-time.”

In addition, every journey a car makes autonomously will be different. Even if the same roads and routes are driven repeatedly, weather conditions will change, traffic density can alter, visibility will vary, etc. Thus, an AV platform needs to learn from these experiences, so it can continuously improve its driving abilities.

Until recently, it was thought that the most ideal way for platforms to “learn” was via real-world driving. Companies like Cruise and Waymo racked up literally millions of miles test-driven by their AVs in the pursuit of circumstantial “edge cases” and unpredictable scenarios that might only occur once in 100,000 journeys yet still would prove invaluable for saving lives of drivers and passengers in the future. In order to rack up these real-world test miles, large fleets of AVs and test drivers needed to operate hundreds of days per year, covering all types of roads and driving conditions. Maintaining such extravagantly outfitted vehicles and teams of test drivers doesn’t come cheap.

Slowly, however, a new method of gaining experience for AV platforms began to be utilized. This was the use of computer-based driving simulators where variables such as road conditions, traffic, weather, and other variables could be endlessly tweaked to create permutations of environmental scenarios that had been encountered in the real world. And just as easily, random, novel, un-experienced scenarios could also be simulated. Because simulations can operate at much faster speeds than real-world journeys and can be run in parallel, it’s possible to accumulate hundreds or even thousands as many “miles” driven virtually as those driven in reality.

The capacity to run through truly infinite numbers of scenarios and possibilities has created an industry around AV testing simulation, and even companies such as Waymo are now admitting that some of the most valuable lessons their platforms learn are gleaned from simulators. While some companies like Cruise and Waymo still conduct plenty of real-world testing, other firms such as the UK’s FiveAI and Japan’s Ascent Robotics prefer the bulk of their AV testing and learning to be done via simulators.

But regardless of which method is employed, there are still costs associated with conducting such extensive testing programs. Thus, companies such as Waymo, Aptiv, and others have turned to ridesharing as a way to bring in revenue as the slow journey toward the achievement of universal SAE Level 4 and 5 functionality progresses.

In the beginning: a brief history of ridesharing

Ridesharing initially was an idea based around greater efficiency that geolocation services—delivered through and by smartphone apps—could provide. As opposed to calling a taxi service whose dispatcher might have to mentally calculate which of his vehicles was closest to you, your smartphone’s location services could perform this calculation for you and send a signal to the vehicle’s operator, all without anyone having to speak over the phone or look at a map.

Very quickly, it became apparent that this intelligent dispatching was much more efficient and could impact trip pricing, particularly if travelers shared their rides with other app users. The first company to capitalize on the idea of ridesharing was Uber, and its tech-savvy founders didn’t stop with mere efficiencies realized from intelligent dispatching and shared rides. Uber was uncompromising about examining every possible cost related to its services, from auto maintenance to insurance to fuel, and what it quickly found was that the human driver for each Uber ride was the biggest cost by far for each trip.

With the auto industry then actively beginning to pursue automated driving, Uber rapidly realized that by eliminating the driver, ridesharing services could realize as much as an 80% increase in profits from operating a fleet of fully autonomous ridesharing vehicles. The term “robo-taxi” began to be used sometime before 2013 to describe ridesharing services that would one day be based around SAE Level 4 or 5 AVs.

From ridesharing to robo-taxis

It took until 2015 before the first true robo-taxis appeared, in the island nation of Singapore of all places. Massachusetts Institute of Technology (MIT) spinoff company nuTonomy operated a small fleet of Mitsubishi i-MiEVs and Renault Zoes in limited areas of the country before being purchased by auto parts firm Delphi (later renamed Aptiv). In 2016, Uber began operating a robo-taxi service under its Advanced Technical Group (ATG) banner to test users in Pittsburgh, using a fleet of specially modified Ford Fusion sedans. Uber then tried operating a fleet of Volvo XC90 AVs in San Francisco before relocating them to Tempe, Arizona where the company experienced the tragic incident described above when pedestrian Elaine Herzberg was killed by one of its AVs. This caused the company to temporarily suspend its robo-taxi services and cease operating AVs in Arizona.

For its part, Aptiv continues to operate limited test robo-taxi services in Singapore, but in the meantime, the firm has partnered with ridesharing company Lyft to offer SAE Level 4 robo-taxi service in designated mixed-traffic areas of Las Vegas. While these robo-taxis technically still have a backup safety driver behind the steering wheel, Aptiv continues to actively make progress toward fully driverless versions of its AVs. Aptiv has also announced it’s forming a joint venture with South Korean automaker Hyundai to create an AV platform for robo-taxi services, fleet operators, and international automakers.

Elsewhere in the U.S., the company Voyage was founded by veterans of online learning business Udacity (which offers many courses and programs related to AVs). Voyage currently operates fleets of SAE Level 4 robo-taxis (with backup safety drivers) in low-traffic retirement communities in Florida and California where there are few pedestrians or other cars. Voyage recently signed a partnership agreement with FiatChrysler for purpose-built Chrysler Pacifica AVs that are designed to be driverless.

One enterprise that currently operates a truly driverless AV robo-taxi service is Waymo. In the suburbs of Phoenix, Arizona, the company’s Waymo One business, like Aptiv’s robo-taxi fleet, often employs safety drivers as a backup for its SAE Level 4 AVs. But not always—at least some Waymo One AVs have no human driver aboard. Waymo has said that over time, the ratio of its driverless AVs to those with a safety driver will increase, with a goal of ultimately eliminating human drivers entirely.

For now, Waymo One only operates in the Phoenix area, primarily because of the city’s relatively light traffic and mostly unchallenging road and weather conditions. But simultaneously, the company is engaged in AV testing in the suburbs of Detroit in Michigan, with the hope of gaining enough experience with that locale’s winter weather conditions to be able to offer driverless robo-taxi services in many more cities throughout the U.S. and the rest of the world.

Other robo-taxi efforts

Robo-taxi-specific ventures have been founded and currently operate. These include Foster City, California-based Zoox, which announced plans in 2015 to build robo-taxi vehicles from scratch, rather than modify existing vehicles. It’s scheduled to reveal its first vehicle later this year.

Like Uber, ridesharing firm Lyft is very interested in autonomous vehicles, having partnership deals with Aptiv (see above) and Waymo (which has a major investment in Lyft and at one point agreed to supply AVs to Lyft) for their AVs to be run in Lyft’s network. In 2018, General Motors invested $500 million in Lyft with the intention of deploying robo-taxi AVs in the U.S. in the near future.

In 2018, AV platform maker Drive.ai began offering a test robo-taxi service in Frisco, Texas that it later expanded to Arlington, home of the Dallas Cowboys. In September 2017, Lyft said it would be working with Drive.ai to offer robo-taxi services in San Francisco. But this plan and the robo-taxi operations in Texas were shuttered when Drive.ai was acquired by Apple in 2019.

American automaker Ford acquired Pittsburgh-based AV platform builder Argo AI and is aiming to launch robo-taxi-style services in three U.S. cities using Argo AI technology. The Argo platform will run in a compact AV that will most likely be based on the Ford Transit Connect minivan.

The other major investor in Argo AI is Volkswagen, which intends to use Argo AI’s technology in a robo-taxi-type capacity associated with the automaker’s present MOIA ridesharing business in Germany. Volkswagen had previously discussed using AV technology from California-based Aurora, an AV platform maker backed by Amazon.

AV platform company Intel/MobilEye currently operates a ride-hailing service in conjunction with Volkswagen in MobilEye’s home of Jerusalem, Israel. The plan is to augment the service with robo-taxis by 2022. Intel/MobilEye also has stated that in the future, its SAE Level 4 and 5 functionality will be intended only for robo-taxi services and auto fleet use (as opposed to production cars for individual owners).

European and Asian robo-taxi ventures

In late 2017, French AV startup Navya unveiled its low-speed, driverless, six-passenger Autonom Cab robo-taxi minivan with SAE Level 4 capabilities. The AV was tested initially by global transport firm Keolis and Western Australia’s Royal Automobile Club (RAC). In 2019, Navya began offering the vehicle to mobility service operators.

German automaker Daimler has announced that it wishes to have a fleet of Mercedes AVs that it would like to operate within Uber’s network. The company also began piloting a separate robo-taxi service in San Jose, California in conjunction with German engineering and technology company Bosch.

At the 2017 Geneva Motor Show, Volkswagen showed off the Sedric concept AV, which was designed to be used as a robo-taxi. The same year, Honda showed off its NeuV concept AV, which was designed to be a “personal” robo-taxi.

In Japan, a number of AV companies, including DeNA, ZMP, and Toyota, had stated they had wanted to have robo-taxis operating for the 2020 Tokyo Olympic Games, but the games were postponed until 2021, so those efforts have been delayed. Separately, DeNA had announced it would be partnering with the Renault-Nissan-Mitsubishi Alliance and public transport operator Transdev to offer robo-taxi services at some point in Japan in the next 7 years.

In China, Didi Chuxing is one of Uber’s largest rivals, having purchased Uber’s China operations in August 2016 (Uber still owns 15.4 percent of DiDi Chuxing). DiDi Chuxing currently dominates at least 70% of the Chinese ridesharing market. With significant investments from Toyota and Apple, DiDi Chuxing opened AI and AV R&D offices in Mountain View, California. In August 2019, the AV portion of Didi Chuxing became a separate company. In May 2020, DiDi Chuxing announced it had plans to begin offering a robo-taxi service in Shanghai.

Another large player in the China market is mobility provider WeRide. WeRide started offering test rides (with safety drivers) in robo-taxis in Guangzhou, China in December 2019 as part of a partnership with Baiyun, the largest taxi company in Southern China. Expansion to other cities in China is scheduled.

Yet another Chinese company to watch is search-engine giant Baidu, which has developed its Apollo open AV platform that has been adopted by more than 60 companies. In September 2019, Baidu began a trial of 45 robo-taxis in Changsha, China, with plans for additional vehicles in the works.

Other companies either operating or planning robo-taxi services in China include AutoX (which offers the highest-speed robo-taxis in China), real-estate firm Vanke, and Pony.ai, the latter of which began a pilot robo-taxi service in Irvine, California last year with South Korean automaker Hyundai.