Comprising an area approximately twice the size of New Jersey, the Netherlands punches above its weight, to use a boxing metaphor, in the piloting and implementation of autonomous vehicles. This is one takeaway from the recent TU-Delft webinar, Deployment potential of automated minibuses for first/last mile transport.
Professor Bart van Arem explains that TU-Delft’s mission is to use science and engineering and design to positively impact society. Autonomous mobility has that potential according to van Arem. He showed images of a paved-over, car-centric built-environment transformed into one centered on people. Still, it is not obvious when automated driving will be commercially successful in Europe and what the transition will be to such a future.
To improve sustainability, electric and shared will have to be part of the autonomous mobility equation as well. These autonomous vehicles will have to operate in shared spaces; spaces that are shared with cycling and new mobility modes. The vehicles must be right-sized for the given trip.
Over 100 FMLM Pilots #
There have been over 100 first/last-mile (FMLM) shuttle experiments in Europe. Irene Zubin presented the results of a survey of the experts (PDF)s behind these pilots and summarized the results of her literature review. One of the biggest barriers has been the low operating speed of the shuttle. Simply, bikes are often as fast.
And, the current costs of the vehicles, estimated to be €280k, are not economically viable. One panelist estimated the long-term cost of €65k for an autonomous shuttle. Another barrier is the financial cost of an onboard safety operator.
There are use cases where that safety operator can potentially perform double duty. Joop Veenis of The Future Mobility Network showed an application where the shuttle essentially serves as a combination hospital reception room/parking lot shuttle. In this case, the shuttle’s onboard operator acts as a point of the first contact to the hospital.
Patrick van Norden of MRDH (Metropoolregio Rotterdam Den Haag) spoke of 8 projects that are pilots or in production in his region, including the 3rd generation Parkshuttle Rivium. The latest generation will operate in mixed traffic over an extended route that will connect Rotterdam’s metro network on one end to a waterbus route to the Drechtsteden. It builds on the first and second-generation driverless shuttles that launched in 1999 and carried over 8 million passengers over a 1.8km dual-lane guideway.
This shuttle operates on 2.5-minute headways during peak hours and on-demand during off-peak hours. Its operation is between 6 AM and 9 PM. The second-generation reportedly had one operator for the 6 driverless vehicles. Still, van Norden indicates that, in general, the industry is behind where it needs to be for widespread deployment. He describes the situation as being
“A little bit in the valley of despair. [Transit agencies] want to buy the futuristic vehicles, but the automakers aren’t as far along as they want. The business case needs to be driverless. Wants only one or two supervisors per 8 vehicles. They are going the wrong way.”
Joop Veenis listed 13 challenges (see slide 15 PDF) – ranging from technical to price to operational to legal and public acceptance – that are hurdles to mass autonomous shuttle deployment.
Integrating the Built-Environment & Other Modes #
At the same time, Veenis suggests that mobility automation is more than delivering people via traditional roads. Some examples include:
- grocery deliveries via right-size devices,
- self-driving garbage cans
- an automated ferry that is set to launch this summer.
Automated mobility will have to work with multiple modes of transportation, including bikes, in Veenis’ vision.
Veenis colleague, Ahmed Hashish, gave an overview of the 2020 Helmond, Netherlands pilot.w This pilot operated in a 3 km long route through mixed traffic stopping at four bus stops.. One of the things that made this particular pilot unique is the metadata integration of the service into Google maps.
This was one of the more integrated pilots of the recently completed, European Union-funded FABULOS (Future Automated Bus Urban Level Operation Systems) project.
FABULOS consisted of six pilots that tested autonomous vehicles in real-world deployments. Still, like the other pilots, the common finding is that more work is required – from a technical, environmental, regulatory, and market perspective – to turn these “Pre-Commercial Procurements” into everyday extensions of public transport.
Automated Mobility District Aspirations in Austin #
Dr. Kara Kockelman presented joint research with Yantao Huang of the concept of adding Automated Mobility Districts (AMDs) along a portion of Austin’s 32-mile commuter rail service. In their model, SAVs (Shared Autonomous Vehicles) act as robotaxis for the first and last-mile access in a confined/geofenced neighborhood.
They used the SUMO Simulation of Urban Mobility tool to simulate agents (the travelers and the Shared Autonomous Vehicles – SAVs) traveling to and from five Austin train stations. They assumed
- the year 2030 forecast for the travel demand model
- used 246 traffic analysis zones (TAZ) of the 6-county region’s 2,252 TAZs.
- Vehicle routing plans were generated dynamically every minute based on the ride requests and their respective locations.
- The SAVs provide dynamic ridesharing and the first-mile and last-mile are provided in one routing plan.
Unlike the lower speeds of the European pilots, 45 MPH was the maximum speed of the simulated SAVs and the average ride distance was 2.52 miles (which includes a 1-mile detour on average because of ridesharing). The assumed SAV fares were one dollar per mile, with railway headways of 15 minutes and ten percent of the travel demand simulated. Some of the results from the simulation include:
- Faster speeds than walking (15 minutes average from the time a rider requests service to drop off).
- A 3.6% reduction in Vehicle Miles Traveled, assuming the riders had been traveling by car.
- Transit Mode share increases 1,000%, from 0.4% to 4.1%
One of the more interesting findings is that the optimum vehicle size is four seats, which reinforces the comment made by van Arem about having the right-sized vehicle for the given application.
Future work includes modeling more realistic congestion, optimizing SAV parking locations, and optimizing the geofence. Kockelman and Huang’s existing research seems like it could provide a good baseline for some real-world testing, similar to what Valley Metro and Waymo did in Phoenix.