Why We Created Bestmile: Commercializing AVs is Harder than Teaching Them to Drive
April 6, 2020 | Company Blog
Safe, reliable self-driving technology is proving stubbornly difficult — harder than prognosticators though just a year ago. Early this year AV tech maker Argo AI CEO Bryan Salesky said driverless cars that could go anywhere are “way in the future.”
While the focus of the industry has been on autonomous vehicle technology, another challenge in bringing AVs to market is the commercialization of the services that they might offer. “There is a growing recognition that getting the self-driving algorithms right is merely an entry ticket for the much bigger challenge of commercialization,” Financial Times writer Scott McKee said.
McKee interviewed former Waymo executive Shaun Stewart who told him that deploying autonomous services at scale is, “a huge feat — it’s more difficult than landing people on the moon.” Dan Ammann, head of GM’s autonomous Cruise business, said the company postponed a fleet of driverless cars when the company realized that the large scale deployment would be “10,000 times harder” than teaching a car to drive through traffic.
This is the challenge that Bestmile was created to solve. We founded the company after we were asked to manage one of the first demonstrations of autonomous vehicles in Europe. The vehicles were programmed to follow a programmed route and detect pedestrians and obstacles, but they couldn’t work together as a fleet to provide a service. Something had to tell them were to go, when to go, how to get there, and where to go next.
Delivering such instructions to a single vehicle might seem simple enough. But consider a city like Chicago, which has some 2,700 taxis that provide an average of 30,000 rides per day. Automating the dispatching of thousands of vehicles as they respond to tens of thousands of ride requests in real time represents and enormous challenge. There are many variables that need to be considered, and the amount of variables grows exponentially with the size of the operation.
The number of variables that must be considered is staggering: the locations of all travelers and all vehicles, the destinations of the travelers, traffic, energy or fuel levels in vehicles, vehicle occupancy capacity, and where the vehicle can go after dropoff to find a new rider as soon as possible. The challenge becomes even greater when rides are shared, which most people expect to be the predominant use case for autonomous mobility services. Shared, or pooled services combine rides for travelers with similar destinations.
All of these calculations and assignments must be made against the backdrop of operator business requirements. To create a viable business, autonomous vehicle service providers will need to offer a reliable service with predictable wait times, ride times, and excess ride times when sharing or pooling rides. They will also need to maximize vehicle utilization — -the time and miles/km spent with paying passengers onboard in order to minimize so-called deadheading, or the time vehicles are cruising while empty.
Can it be done? No service provider has deployed a fleet of autonomous vehicles at this kind of scale, but Bestmile conducted a test of sorts using the demand and ride data from Chicago’s 2,700 taxis and an average day of 31,000 rides, traveling 78,000 miles. Using the optimization algorithms we have developed for our Fleet Orchestration Platform to enable mobility providers to turn autonomous and human-driven fleets into viable businesses, we created a virtual fleet of shared vehicles to test the performance of a Bestmile-optimized fleet with the taxi demand and trip data as a benchmark.
The results were very promising. Our simulation of a day in the life of Chicago’s taxi fleet found that just 400 shared vehicles could handle the entire demand, traveling just 46,000 miles, with an average occupancy of more than three people per vehicle and marginal excess ride times due to ride sharing. We could control the ride times and wait times by adjusting the fleet size. The gains from the fleet orchestration technology — fewer miles traveled and vehicles used, higher occupancy rates — are measurable and significant.
It should be noted that the results of this simulation in the city of Chicago could be delivered by autonomous or human-driven vehicles. The difference would be that the Bestmile Platform’s automated dispatching would either issue instructions to drivers via a mobile app, or directly to a vehicle’s onboard self-driving tech stack.
Advocates for shared autonomous mobility hope it will result in reduced urban congestion and pollution. Of course, getting there requires that vehicles are able to drive safely in a variety of conditions. But it also requires that the services are able to efficiently move more people with fewer vehicles, and that the services are economically sustainable. This second challenge has not been at the forefront of public discussion about autonomy, but has been the focus of Bestmile since 2014.