Study: AI and Driverless Cars Could Slash Taxi Fares in Austin, Texas
The model uses software agents in a network; the agents represent robotic taxis, and the network stands in for the traffic system of Austin.
…Fagnant and Kockleman made their simulated robots wait at the curb for just five minutes, and they set all the trips to begin and end at points that lie inside one of the 2,258 zones into which Austin’s traffic is divided.
They used real-life commuting data to simulate the four rush periods of the day, and they calibrated the size of their robotic fleet in a preliminary simulation that asked how many cabs had to be available on the first day to ensure that no traveler had to wait long for a ride. The answer: 1,715 robotaxis, providing 56,324 person-trips a day.
To get efficiency and fairness alike, the researchers juggled a number of variables, not least the convenience of staying in a taxi nonstop versus the efficiency of sometimes being handed off to another one. They gave priority to those already in a taxi, and they also made sure that no traveler who’s almost at his destination must suffer through a late-minute detour to pick up someone else. Finally, the researchers focused on a small sample of riders meant to represent the one or two percent of all travelers who would be early adopters.
For this sample, they found that every robotaxi replaced about 11 cars; the average wait for a taxi was a bit over a minute, with 99 percent waiting 10 minutes or less. Because the early-adopter sample was small, relatively few people shared rides. But with broader use, that percentage would rise, allowing still greater efficiencies.