The Lost Generation of Automated Driving

May 4th, 2022 | by Andreas Richter

(15 min read)

In 2016 the era of autonomous driving projects with customers under real-world conditions started. More and more initiatives popped up to test how autonomous bus shuttles work in reality and the trend hasn’t stopped, yet: taking one or more automated minibuses on a predefined route (often on private ground) and starting a few kilometers of low-speed service.

We asked ourselves: What have we learned from so many projects doing the same? And how does this relate to all the current SAE Level 4 automated driving activities? Let’s do a non-exhaustive research.

Who will win the battle of SAE Level 4 driving? Minibus companies or robo taxi developers? (Wiener Neustadt, image by Marius Dupuis)

Minibus Players in Europe

If one is located in Europe, the most prominent self-driving minibus players are EasyMile and Navya, both French companies with around 250 employees each. They already have deployed minibuses for more than 300 projects world wide. They have in common that both companies provide an all-inclusive solution: Starting with a feasibility study and checking the local regulations, the route will be mapped and the staff trained. After successful testing without passengers, the service can be launched 24/7 supported by the companies’ fleet operation management tools. Their vehicles can carry between 12 and 15 people and could go with up to 50 km/h… but often they are limited to less than 25 km/h. To be honest we didn’t find any project with a higher operation speed.

Technical Challenges

If you only look at the projects in Germany (60 up to now!), you can have the feeling that all projects would like to do the same: Deploy the “next generation” “innovative” “disrupting” solution of mobility as a service for urban (e.g. in Magdeburg) and rural areas (e.g. in Mansfeld-Südharz) for first/last mile as well as on campus sites (e.g. in Kiel). Test it in real-world condition, see how it works with vehicle-to-any-communication (V2X), develop operations and charging management, test user experience and acceptance and see if this whole stuff is economically viable. Luckily (or surprisingly) the results are more or less the same in all the projects: The buses are too slow and they drive way too defensive. They are slowing down the other traffic participants (even bicyclists). The faster the buses drive, the more clearance they need which leads to more emergency braking when other road users pass by (even opposing traffic, crossing cats or grass moved by wind (a comprehensive list can be found in a chapter about the Bad Birnbach project)). Due to the limited course the buses are not able to deviate their routing to react on blockades and become an obstacle for other road users because they cannot reverse on their own and stay stunned where they are. In videos you often see tire tracks on the ground because the vehicles are driving exactly the same course every time! Sometimes, only for these buses road marks had to be applied (like in Bad Birnbach). Bad weather conditions such as rain and snow as well as strong wind or fog is blinding the vehicles which led to the final cancellation of the test in Vienna’s Seestadt or protests in Berlin. Often, an operator has to move the bus with a controller but in this case the speed is limited to only 5 km/h. The operators are currently a must-have on public roads and they have to be trained and paid which is currently not a success story for being cost-effective.

To be fair, the low speed is not a homemade problem of the self-driving mini buses. As already mentioned, they could technically go up to 40 or 50 km/h but there are currently no regulations that would allow an insured operation. Therefore, the tests are often placed on private grounds such as campuses or in low-speed zones when it comes to mixed traffic situations. In Bad Birnbach it got even more complex because for some meters a rural road had to be used. The solution is to reduce the speed difference between bus and other traffic participants: Cameras detect the approaching bus and variable message signs are set to a general speed limit of 30 km/h for all traffic.

User Experience

Interviewed users often say that they like (here an example from Canada) to ride on such vehicles. When asked, most reveal that they do this for the first time and only because they want to give it a try. They like that the bus is driving autonomously and that the safety driver has nothing to do (may be pushing the button for the ride to continue after calling at a bus stop. But while riding disillusionment follows swiftly. The speed of the buses is quite low. They stop quite often (sometimes hard) in situations where the passengers do not see any real risk. These safety maneuvers reduce the trust of the riders because they don’t understand why the bus is interpreting the situation differently from them. And the riders are upset when the bus is not in service because of harsh weather conditions (which is the best time for sheltered public transport instead of a stroll).

But the passengers are not the only users. The operators are also experiencing these buses and realize that their maintenance is also not trivial and can get expensive. An accident (even a car body damage) may lead to high costs because of affected sensor pods. Currently, LiDAR sensors can cost as much as a compact car.

Also the update of the software has to be done by an OEM professional with LAN access. The process costs some hours with additional calibration and it is not guaranteed that all situations currently handled successfully will be handled the same way after the update. Thus, additional tests have to be applied, too. If you then take into account that you need specially trained staff for operating as security driver, the operating costs are getting even higher. No public transport operator is thinking about including such service into production, yet.

In general, the routes of current projects are too short for a meaningful usage in daily commuting. The above mentioned drawbacks in addition lead to the conclusion that these automated minibus services are currently just like a tourist attraction but cannot really add value to the overall (public) transport problem. One can summarize that these buses are working more or less in SAE Level 2 or 3. It seems that there is a long way to reach Level 4. Nevertheless, these buses are interesting tourist attraction and may increase the number of visitors.

Minibus Players in the Rest of the World

But there are other companies working on self-driving minibuses than only EasyMile with their EZ10 or Navya with their Arma (they also offer a more cab-like vehicle called Autonom Cab). A comprehensive list including prototypes and concepts can be found in IDTechEx’s report on Robot Shuttles for Smart Cities 2021-2041. The most prominent examples are:

  • Toyota had their e-PALETTE (Video) (don’t mix it up with the EUR-pallet). It was used during the Olympic Games 2021 in Tokyo but not nearly on the scale it was intended to. The service was discontinued after an accident with a pedestrian (in manual mode at 2 km/h). It is planned to use the platform in Woven City, Toyota’s Smart City testbed, for all kind of transport.

Additionally, Tier-1 suppliers are also working on such mobility solutions:

  • For example, iav has also an own AD bus platform, currently deployed in Hamburg or used for internal tests by Siemens in Munich. The vehicle in Hamburg (video) is also slow-moving on a short course and needs vehicle-to-infrastructure-communication (V2I) to get the state of the traffic lights. All traffic signs have to be implemented in the map; there is no road sign recognition.
  • On the one hand, while iav’s bus is still on prototype level (but at least in the HEAT project planned to go up to 50 km/h), ZF’s (or originally 2getthere’s shuttle, the company 2getthere from The Netherlends was bought by ZF) GRT automated minibuses are already in public service. On the other hand, these systems are often driving on separated lanes like an automated people mover system. Nevertheless, there is the plan to test the ZF shuttles in Friedrichshafen with higher speeds to keep up with traffic on a rural road.
  • The Tier-1 supplier Magna recently bought Optimus Ride to strengthen their advanced driver assistance business. They build a rubber-wheel-based tourist-train trailer looking like a vehicle for up to six persons (video) derived from the Polaris GEM golf carts. It is also designed for SAE Level 4 and low speed. Operations were conducted mainly on private grounds. These golf carts are also used by May Mobility to add wheelchair accessibility to their Mobility-as-a-Service (MaaS) solution.
    By the way: These Polaris GEM golf carts are also used by Perrone Robotics and Ford for testing automated driving. But remember, Waymo started also in the golf cart class with their Firefly vehicles…
  • The French Lohr Group, together with operator Transdev, starts to automate their i-Cristal minibus (video). This is an electric vehicle with up to 7 seats and a top speed of 50 km/h. The interesting detail is the concept to connect more minibuses to a road train to extend capacity. The self-driving system is provided by the Intel company Mobileye. This is part of Mobileye’s strategy to deploy their SAE Level 4 system called “Drive” into different vehicle platforms for robo taxies, shuttle buses for Mobility- and Transportation-as-a-Service (TaaS) as well as private cars.
    Mobileye also cooperates with Tier-1 supplier Schaeffler to use their rolling chassis solution to build MaaS/TaaS vehicles and automate them with Mobileye Drive.
  • The US-Philippine company COAST is working on a little bit larger kinds of shuttle buses: The P1 (videos) is the hardware for the turnkey Mobility-as-a-Service solution (including mapping, routing, fleet management, customer interface, etc.) they would like to provide – for private sites. Unfortunatel,y no further details were published.
  • In Taiwan the local industry also collaborated to build an automated minibus, the so called WinBus (video). It is also foreseen to reach SAE Level 4 but until now tests were only conducted on a 7.5 km long, fixed-route track with at least some interesting scenarios such as unprotected intersections and lane switching. The route will be extended to 13.4 km and it is planned to operate at 50 km/h.
  • In New Zealand local partners started a self-driving minibus project with Ohmio to aim for SAE Level 4. First, they started in Christchurch with the HOP for six people and continued the development to the LIFT (video) for 20 passengers. This vehicle was already selected to conduct tests outside Christchurch by Lotte, one of the biggest conglomerate corporations in South Korea with the long-term plan to create a fleet of up to 250 buses.
  • The Tallinn University of Technology (TalTech) and Auve Tech developed an eight seat bus shuttle with SAE Level 4 capabilities (video) called Iseauto. The maximum speed is specified with 30 km/h but on a short part 50 km/h are supposedly reached. The Initial idea was to build a bus system that will operate outside the rush hour not only on predefined routes and schedule. At least they tested their vehicles also on snowy and unpaved roads. And, together with the University of Tartu, Auve Tech introduced a hydrogen fuel cell to the shuttle buses to achieve more service hours.
  • In Stockholm the Swedish-Chinese car manufacture NEVS also would like to set up a bus shuttle ecosystem (called PONS). Their vehicle Sango is a six seater and currently driving at 15 km/h but should later accelerate up to 50 km/h (video). Of course, it will come with SAE Level 4 capabilities that interestingly are provided by the Chinese Level 4 company AutoX. A test with ten vehicles was to be conducted in Stockholm but didn’t really take place. Maybe this is related to the financial embarrassment of the owner Evergrande.
  • In Helsinki (at the NOKIA campus) you can find the Gacha. It is also a shuttle-like bus for up to 40 km/h but it is bigger than the other ones. Up to 10 people can find a seat. At least it was presented in contrary weather but nevertheless was able to take the demonstration rides (video) and, indeed, the Finnish developing company Sensible 4 is focusing on all weather conditions. The company is not limited to their own bus design and recently started a collaboration with the German Moove GmbH to build a larger people mover as well as a cooperation with the Chinese Manufacture Dongfeng for targeting mass production.

Talking about China there are also bus shuttle activities conducted.

  • One of the best known companies related to automated driving is Baidu. They are not only investing in their open-source platform Apollo for automated driving and working on SAE Level for Mobility-as-a-Service solutions (with the Apollo Moon vehicle built by BAIC) but also started to produce mini buses (called Apolong) in a larger number, which are built by Chinese bus brand King Long. A small King Long bus with 13 seats was also adapted to work in autonomous mode (video).
  • The Chinese bus brand Yutong (don’t mix it up with the autoclaved aerated concrete bricks from YTONG) is also investing in automated bus shuttles. The Xiaoyu named bus is produced in a larger number. Information about this type is rather limited but one can assume that it is also operating with max 40 km/h (video). And, as often in China, test sites are equipped with V2I communication technology (using 5G) so that the vehicles can easily anticipate the status of the traffic lights.
  • What else do we have? Another Chinese bus brand Zonson presented their slightly smaller shuttle called BAZN (video). Also SAE level 4, also low speed, also limited information available.
  • And the biggest Canadian bus brand New Flyer (NFI Group) is working on automated driving in heavy duty size (up to 52 seats). An Xcelsior bus was equipped with the usual amount of sensors for SAE level 4 driving. The self-driving system itself is developed by Robotic Research. Unfortunately, no information was found on their targeted speed but at least we can assume that the Xcelsior class is heavily tested regarding their crashworthiness because it is a retro-fitted model with long operational experience.

Crashworthiness of the Mini Buses

By this time we know that automated driving vehicles are rarely causing accidents. Often, they are prevented by the safety driver (esp. in the SAE Level 4 domain) but we also already read about the very defensive way of driving of these mini buses. But they are not spared from the insufficiency of other traffic participants. Fortunately, often only so called “fender bender” accidents happen. Apart from this, one can only find few information about the crashworthiness of the vehicles which look kind of fragile. EasyMile has published some information about their EZ10 in a safety report. There you can read that the threat of rollovers is reduced by the weight of the vehicle and its low speed and acceleration. But what happens if a third-party vehicle is slamming right into the shuttle? The only answer you can find is that the EZ10 is designed with a robust aluminum roof structure and steel for sub-structures to protect passengers in case of a rollover.

Local Motors at least published more information about their 3d-printed Olli. It is able to withstand a crash hitting a wall with 40 km/h. Windows shattered but the main structure was not critically impacted. In real-world operations the Ollis do have laminated glass so shattering will not occur. But now Local Motors stopped the business and from all the other players there is no such information available (and if it is, please add it to the comments!).

Takeover of Level 4 Players

Zoox and Cruise already have shown their mini bus concept and they are using their SAE Level 4 driving stack for bringing them onto the road. Volkswagen would like to use their electric mini bus for Mobility-as-a-Service solutions using Argo.AI’s self-driving system focusing on European and American cities. Also the Chinese players such as AutoX, Pony.ai, WeRide, Momenta as well as Baidu work on automated driving solutions to be used as robo taxis (or deploying their robo taxi solutions to robo buses).

All their focus is on deploying their system on general roads (from town highways to local residential roads) with corresponding speeds (e.g., up to 80 km/h), handling a large variety of road participants (e.g., trucks and breakneck driving scooters), various weather conditions (such as rain and snow) and especially not on predefined routes but rather on-demand within a geofenced service area. 

Adverse Weather Capabilities

The question is: Why to invest in the small mini bus companies providing low speed solutions in very limited operational zones struggling with forceful road participants and bad weather when there are players out there with 10-fold number of developers and huge vehicle fleets collecting thousands of real-world and simulator miles every day. Some of them have already shown that they can handle adverse weather conditions like Pony.ai or drive on snowy roads like Yandex. But to be fair: A lot of these tests are conducted in locations with good weather, good road conditions and separated traffic, less traffic density and therefore only little congestion, more or less traffic-rule-compliant road participants and, especially in China, with massive support of vehicle-to-infrastructure communication. Thus, demonstrations could look more promising than they really are. And at least in Germany yet another shuttle bus project is starting but this one is explicitly focussing on how current sensors and software can guarantee all-weather capability. Its called “KelRide” and uses an EasyMile EZ10 to create a “Weather-Proof Smart Shuttle”.

Disengagement Rates

The power of the level 4 players can clearly be discerened while having a look into the latest Disengagement Reports from California: Waymo has nearly 700 vehicles on the road, followed by Cruise with roundabout 140. You find Zoox with 85 vehicles as well and Pony.AI with nearly 40 – and there is also EasyMile with exactly one vehicle.
Waymo drove almost 4 million kilometers with roughly 13 thousand kilometers between disengagements (they got worse because of expanding to new testing areas and they started using the Jaguar iPaces). Cruise reached more than 67 thousand kilometers without a disengagement and Zoox is very close to Waymo with roughly 12 thousand. Number one is AutoX with more than 80 thousand kilometers without any disengagement and, interestingly, this is the same amount of mileage of their whole fleet (44 vehicles – in China they piled up to 1,000 robo taxies recently). In contrast to all of the previously mentioned level 4 players, EasyMile manages only 515 kilometers before the system had to be disengaged.

To be honest, looking at California is not looking at the rest of the world. EasyMile and Navya have a lot of projects running around the world. But the disengagement reports with the number of vehicles and mileage can give you a good impression. However, the question why to deal with and invest into small mini bus companies is still unanswered.

Working with Municipalities

One answer is that with these mini bus companies, small municipalities can work on a cooperation level whenever the big level 4 players try to protect their intellectual property as much as possible. Working together in projects like EasyMile and Navya are doing, would unveil too much information about challenges and situations. Waymo recently tried to get rid of the reporting need of disengagements in California to avoid giving competitors insight about their development challenges. Also, the pick-up and drop-off maneuvers of Cruise in San Francisco caused the public authorities to have a look into it because the vehicles are not handling the curb as they should do.

With small-size projects it is easier to include local industry and research; more stakeholder can participate and it pays off for them somehow. Such projects can easily be deployed to rural or remote areas whereas the big level 4 players are interested in implementing their services in metropolitan areas where a larger amount of mobility demand is located. And it seems that there is actual demand (not only because of the more than 60 similar projects mentioned at the beginning of this feature) because EasyMile recently established a new service hub in Berlin to be able to maintain the growing amount of their buses and provide a better service to their customers.

Conclusion

Maybe, at some point the SAE Level 4 self-driving systems are so good that they will have no issues in rural areas if they are already able to handle complex urban environments. Maybe not, because these self-driving systems still rely heavily on their own map representation (but we learned that also the mini bus providers need comprehensive map data) and therefore these systems cannot really scale. Recently, Navya partnered with Valeo to strengthen their safety system development to keep pace with the big level 4 players. Also, a collaboration with bus manufacturers can lead to a reduction of costs by using their vehicle platforms for automated driving. Other players try to implement their self-driving stack to robot taxis such as May Mobility or Sensible 4. But it could also be very likely that the small companies will stop their work, like Local Motors, or get acquired by Tier-1 suppliers and only the big players with enough resources will remain.

If you want to read more about the pros and cons of shuttle bus applications then have a look into Brad Tempelton’s article about the worthiness of robo-shuttle applications.

One Comment

Leave a Reply to Andreas Richter Cancel reply

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