Then I use a for loop to loop through different cars (that have moved) and for each car, I loop through different trips (most likely, a car will have multiple trips). For each trip, I can get the route info from goole map api (library(ggmap) does it). It also gives the distance between each turn. The route info is an approximation to the actual movement of a car. Car2go doesn’t supply realtime GPS info when the car is moving, it only records when a customer checks out a trip. However, I think the approximated route is close enough to a real life scenario, which assumes most car2go customers use the service for transportation purpose, other than ​leisure and recreation activities.

Then I was able to plot the route for each car using the route data return by google maps. The result is shown on the left below. While it roughly represents the route on the map, it fails for curvy roads with less turns. The reason is that route(output = ‘simple’) only gives instruction for each turn, and between each turn, geom_path uses a straight line.

​In order to solve this problem, I found this article, which converts polyline from goolge map api with route(output = ‘all’) and outputs (lon, lat) coordinates.
Now the path represents the actual route on the road, as shown on the right plot above.

Next, I plot all the trips happened over Dec 07 13:40 – Dec 08 13:36 (Monday – Tuesday). The covered area is, as expected, similar to the service area of Car2go Austin. 

Most trips are less than 5 miles and 50 minutes. Note there are a significant amount of less than 1-mile trips. While some of them are actual trips by customers, the rest could be noise in the data or moveover by Car2go.

Now, we can take a look at the starting time of a trip during a day.

As shown in the above plot, most trips are for commute (~8am and ~6pm) and very few trips took place during midnight.