Can it build on ubuntu 18.04 melodic?

Hi, you can solve a real valued LP if you only assign one request per vehicle in each time-step (vehicles may have passengers onboard). Otherwise, you need to solve the ILP.

@@amrlab Thank you for replying, professor. Assume we are only thinging about the assignment problem at a particular time (the algorithm 3 in the paper). Even if we assign one request for each car, the problem still contains integer constraints (for ex. \epsilon_{i,j}), and an integer programming solver is stilled needed. I wonder if I misunderstand something.

do you know how to find this paper?

Having no solid understanding of the tech behind your obstacle recognition systems (sensors / cameras etc) I wonder how light field technology might benefit (or not) as an alternative method of spatially mapping obstacles.

Yes, we use real NYC taxi data, which is mostly from Manhattan. Indeed, we only considered trips which start and end in Manhattan. The method could be applied to the larger NYC or other cities.

If you had included all trips, how much would the trips going outside of Manhattan decrease your efficiency?

Can you please share how did you clean the data outside of Manhattan ?

Hello, do you mind telling me that how you know the number of active taxis is 13,586 (In you paper On-demand high-capacity ride-sharing via dynamic trip-vehicle assignment)?

Good point. In the video we run our algorithm to route a fleet of taxis, which serve real requests from Manhattan. Each dot represents one vehicle and its color the number of passengers at that time instance. The dots move as the vehicles move through-out a day of activity. We highlight the path of two vehicles, with pick-ups and drop-offs. The video shows: - How the algorithm works - A comparison for three fleet sizes and for three capacities and for the different days of the week.

C++/python with ROS for the robots, matlab for the simulations