Posts Tagged ‘carshare’

Floating carshare research we should do

I have a small code project dedicated to analyzing floating one-way carshare usage data. It’s called electric2go and is open source.

Floating carshare systems offer cars for rent by the minute or hour. Cars do not have to be returned to where they were picked up, but can be dropped off almost anywhere within a specified operating area and picked up from there by another member for the next trip. This mimics most bikeshare operations, and in a sense provides a taxi you drive yourself.

car2go and Drivenow are the major multi-city operators, with some regional systems like Enjoy in five Italian cities, Communauto Automobile in Montréal, and Evo in Vancouver.

I have previously done a bit of statistics and some availability map animations for selected cities. But there’s a lot more that can be done. Here are some of the possible questions for further study and research.

This post is equal parts roadmap, to-do list, motivation for me, inspiration for others, and a call out for the interested.

Usage statistics

There are some obvious statistics: where is the service used most – densest cores, spread-out suburbs, in-between? When are trips most frequently started? What are the differences in use in mornings vs evenings vs late nights; on weekdays vs weekdays? What is the median parking duration, how much of total parked time do parking periods over 12 hours comprise, where are the cars parked the longest, vs the shortest?

How can we quantify availability of carshare vehicles? One idea could be to calculate area (in square kilometers) of a neighbourhood/borough; for any given time, calculate area within 400 m of a carshare vehicle and divide the two to get a ratio; then repeat over time to get graphs of when the given area is has best carshare availability.

From quick analyses, in Canadian cities, floating carshare usage is correlated with population density. But I think there is more. For example, is one-way carsharing more popular in cities with a grid system of streets vs radial patterns? In bigger cities or smaller? Is there an optimal population density for this kind of carsharing? It might not work very well in Hong Kong. Can we model and predict usage of the systems, how long a car will be parked, where it will move next?

Carshare is fundamentally about transportation, so compare it with other transportation methods. Compared to a car parked outside a person’s house, how much trip time does walking to a carshare vehicle add?

Compared to a transit trip, how much time is saved, and on what kind of trips does carshare save the most time? Pick out a sample of trips and run its start and end points through a transit planner specifying time and day of week, to see how driving compared with transit. Which carshare trips – which origin-destination pairs, which times of day – save the most time, and which ones overlap with transit the most?

Are floating carshares used as “last-mile” transportation, travelling to and from rapid transit stations? Could they be – how much time would that save? (E.g.: Calgary/Montréal/Vancouver, outside of downtowns: how many trips between 7 and 10 am end within 200 m of a rapid transit station?) How does the total trip duration and cost compare with using a traditional taxi?

Systems other than car2go often have a choice of vehicles and car2go has started introducing various size 4 and 5-door vehicles as well. Compare the usage of 2-door vs 4-door car2go vehicles; is car2go used more now that they’ve introduced the option? How does usage of 4-door car2go vehicles compare with vehicles from competing systems? In cities with multiple systems (e.g. Milan has four currently!), how does each system’s usage compare and why is it different?

Looking farther into the future, try to estimate how many fewer cars would be needed if vehicles could reposition themselves. Driving back to a busy area at peak hours is an obvious application. However, it would be equally interesting to quantify if and how last-mile connectivity in less dense areas could be improved if a car can move itself slightly (say 1 km) to be closer to the hailer.

Maintenance and operations

Much has been written about plans for transportation services based on self-driving cars. I believe that much of motivation for the car manufacturers running carshare programs is hedging and research for such a possibility. Through their efforts, we can also learn what might be required.

A key of carshare operations is that there is no central base where vehicles are returned and could be maintained. More important than rare repairs or yearly maintenance are everyday tasks required for cars, particularly intensively used cars: cleaning and refueling/charging. Although users are expected to keep the cars relatively clean, over dozens of rentals dirt will inevitably accumulate, especially in rainy or snowy weather. Accidental spills and sickness happen and must also be cleaned up. In a floating carshare, vehicles are also sometimes relocated by the employees.

The maintenance requirements of a self-driving fleet would be fairly similar to that of a floating carshare. It will be useful to see when, where, and how maintenance is currently done.

For an example implementation, pick around 100 cars within a fleet at random. Over one month, chart their cleanliness ratings and fuel level as a scatter plot: x axis = time since last clean/refuel, y axis = fuel level or cleanliness rating. We’ll see the cleanliness spike and the fuel go to 100% when they are cared for. By measuring time between spikes we can determine how much maintenance a car needs.

For finding relocations of vehicles by staff, scatter-plot time spent parked vs hour of day when it was finally moved. I would expect to find a specific time of day where a lot of the long-parked cars are moved, for example 11 am as it is during work hours and should have relatively low road traffic.

There is, of course, a lot more; I also have a technically-oriented list. Contact me if you’re interested in any of the above!

Montréal one-way carshare comparison: Auto-mobile and car2go

Montréal is quite possibly the only city in U.S. & Canada to host two one-way roaming carshare operations, allowing direct comparisons between different companies’ operations.

Communauto, Québec’s home grown carsharing company, was founded in 1994. In June 2013, they started Auto-mobile, a one-way roaming service. car2go became the second one-way carshare in Montréal in November 2013; I compared their operations in Montréal with car2go systems in other cities in Canada in an earlier blog post.

In late March 2015, car2go had around 330 vehicles in Montréal, all gasoline smart fortwos. Auto-mobile had around 140 vehicles, mostly hybrid Toyota Prius Cs (~85%) with a few all-electric Nissan Leafs (~15%).

The two systems’ operating areas are mostly similar. As of March 2015, both systems include Le Plateau-Mont-Royal and Rosemont–La Petite-Patrie boroughs in east Montréal, and much of Côte-des-Neiges–Notre-Dame-de-Grâce and Sud-Ouest in the west. Auto-mobile additionally includes Verdun, while car2go includes a couple outlying parking spots and a few more mainly industrial areas that Auto-mobile excludes. The map below shows car2go operating area marked in blue and Auto-mobile’s in green, with the area covered by both systems in a darker green.

Map of operating areas, the same as discussed in text

I took a look at the use of the two systems during the week starting early morning Monday, March 23rd and ending early morning Monday, March 30th.

Auto-mobile’s total utilization ratio for this period was 22.8%, meaning the average car was used for about 5 hours 30 minutes per every 24 hours. car2go’s utilization ratio was 18.9%, meaning the average vehicle was used for about 4 hours 30 minutes per 24 hours. This includes nighttime when most cars are not used; a private car used for a long, one-hour-each-way commute and then some errands might clock up 3 hours per day, and most are used well less than that.

The use pattern for the two systems is somewhat different. Auto-mobile cars make roughly half the amount of separate trips that car2go vehicles make. The median Auto-mobile car makes 2.93 trips per day, and the 25-75th percentile range is 1.71-4.11 trips per car per day. The median car2go vehicle makes 5.86 trips per day, and the 25-75th percentile range is 4.86-6.71. The most-used Auto-mobile car made 43 trips in the 7 day data period (average of 6.14 trips per day), and the most-used car2go vehicle made 72 trips in the 7 days (average of 10.29 trips per day).

Distances per trip are largely similar (see below for note about how distances are measured). Auto-mobile’s median distance per trip is 1.31 km, while car2go’s is slightly higher at 1.59 km. 10.9% of Auto-mobile trips and 9.9% of car2go trips are longer than 5 km. Although the east-west (Parc Angrignon to Jardin botanique) span of the operating areas is around 16 km, there were very few longer trips: only 0.71% of Auto-mobile trips were longer than 10 km, and for car2go this figure was only 0.20%.

Auto-mobile users make more relatively long trips. The Auto-mobile median trip duration is 32 minutes, and the 25-75th percentile range is 17-67 minutes. car2go’s median trip duration is 25 minutes, and the 25-75th percentile range is 15-33 minutes. 16.6% of Auto-mobile trips are over 2 hours, while the same figure for car2go is only 3.8%. Similarly 6.2% of Auto-mobile trips are over 5 hours, compared to only 1.7% for car2go.

Based on Auto-mobile’s longer trip durations and shorter apparent trip distances, it appears the service is more frequently used for longer round trips with the vehicle being returned near where it was picked up. This might be due to Auto-mobile’s operation as a complement to Communauto’s traditional station-based carshare system. Communauto members are able to use both carshare models interchangeably and might tend to use Auto-mobile cars for longer round trips more frequently, perhaps if an Auto-mobile car is closer than an available Communauto vehicle.

Methodology

The figures in this article are calculated based on vehicle availability starting Monday, March 23rd, 2015 at 4:00 a.m. EDT and ending Monday, March 30th at 4:00 a.m. EDT (8:00 UTC).

This article uses data from the car2go API but is not endorsed or certified by car2go.

For both systems, I was only able to get current availability of vehicles. When collected over time this data allows collation of trips, but some level of inaccuracy will result; for instance, cars that are reserved or blocked are reported as in use by the APIs, so trip time includes reservation time if any. As I only have access to start and end locations for each trip and not the odometer reading, trip distances are calculated as the crow flies. Further, a round trip in which the member does not terminate the trip midway will appear to have a relatively long duration but a short distance.

A week of car2go in Canada

This post has been updated on March 27, 2015 with corrected data. The data posted initially was partially wrong due to a bug in analysis code.

The bug caused trips that started and ended in the exact same position to not be counted. The car2go GPS reports 5 decimal digits (resolution of about 1 m), but I have overlooked that cars at a designated parking spot always show the exact same GPS coordinates predefined by car2go. The bug then ignored trips made by cars picked up and returned to the same designated parking spot.

Depending on the city, this caused about 22% to 33% of trips to be excluded and underestimated total fleet utilization by 18% to 31%. Median trip counts per car were underestimated by 20% to 32%, while median trip distances were overestimated by 22% to 31% (see below for note on round-trip distances). Trip duration median were accurate to within a minute, but quartiles were overestimated by 5% to 10%. Because the bug only affected detecting round trips, the positions maps were accurate and have not changed.

car2go is a carsharing service operating in a number of Canadian cities. It has a one-way model in which cars do not have to be returned to where they were picked up, but can be dropped off almost anywhere within a specified operating area and picked up from there by another member by for the next trip. This mimics most bikeshare operations and in a sense provides a taxi you drive yourself.

Systems are outfitted with a uniform fleet of two-passenger Smart Fortwo vehicles and a parking deal is negotiated, most commonly to allow parking in any resident/permit location.

Data on usage of the service is available from an API, listing current positions of available vehicles. Collated over time, we can calculate a car’s trips, and some statistics about those trips.

I’ve been interested in using this data to create visualizations of the service. I have created a number of animated maps (uploaded to Youtube) that track locations of available cars over time.

This is something slightly different: I took in the data for a week of operations, and extracted all locations of available cars, plotting them on a single map. The result is an indication of use and demand for car2go service within each city’s operational area. Some interesting patterns emerge: correlation with residential and commercial density, impact of parking regulations and age of service.

Composite of unlabelled car2go data for four Canadian cities

(more…)