1. SHARED CAR NETWORK PRODUCTION SCHEDULING PROJECT – SPRING 2014 Tyler Ritrovato (tr2397) Peter Gray (png2105)

2. THE IDEA  Google’s Driverless Car began design in 2005 and continues to advance  Advent of Uber and Lyft services in late 2000’s We see an opportunity… New Driverless Car Technology + Efficient Dispatching Algorithms __________________________ Shared Car Network

3. SHARED CAR NETWORK  Instead of owning multiple cars per household, individuals or families become a member of a Shared Car Network (SCN)  Cars dispatched based on an efficient algorithm BENEFITS  Less cars on road is better for environment  Reduced traffic (at scale)  No more hassle of owning and maintaining personal cars RISKS  Not as flexible for on-demand trips  Potential for late or missed pick-ups

4. RELATING TO A SCHEDULING PROBLEM  Machines  All the cars in the network  Regular Job  Picking up a customer and dropping that customer off. Defined by the following inputs: o Origin o Destination o Pick-up Time o Time due at destination  Processing Times:  Unoccupied Car  time from last drop-off to next customer pick-up  Occupied Car  time from pick-up of customer to drop-off

5. OPTIMIZATION DECISION # of Machines % of Requests Serviced Max Lateness Minimize # of Machines Constrain on Max Lateness and Minimum % of Requests Serviced Therefore, our problem boils down to the following production scheduling problem: P | r j, L max | m

6. SAMPLE DATA  Downloaded September, 2013 data from Citibike.com  Focused on the morning rush hour (8:00 AM- 10:00 AM) on Monday, September 9 th.  Limited data to nine citibike ids (machines)  Release date  Start of trip  Due date  Trip Duration plus 20%  24 Total Jobs

7. ALGORITHM STRUCTURE Utilizing a Greedy Algorithm:  Step 1: List job requests in ascending order (morning to night)  Step 2: For each job, choose the machine with the lowest metric score  Metric Score  Remaining processing time of current job + time to reach customer – time since availability  Add a machine if all of the possible machines lead to an undesirable lateness value  Step 3: Continue until all jobs are assigned

8. ALGORITHM EXAMPLE  Job 1: Starts at 8:01 AM at W 25 St & 6 Ave and ends at 8:12 AM at Broadway & W 51 St  Add job 1 to machine 1

9. ALGORITHM EXAMPLE  Job 2: Starts at 8:04 AM at 11 Ave & W 41 St and ends at 8:33 AM at John St & William St  Must add a second machine because using just machine 1 would lead to being late by 15 minutes  Lateness= 8 minutes remaining processing time from job 1 + 7 minutes to travel from job 1 ending point to job 2 starting point ✓  Add job 2 to machine 2

10. METRIC SCORE EXAMPLE  Job 11: Starts at 9:00 AM at Fulton St and Grand Ave and ends at 9:04 AM at Lafayette Ave and Classon Ave  At this point in the algorithm, there are 5 machines  What machine should job 11 be assigned to? Machine 1: Available since 8:34 and is 24 ½ minutes away from pickup location  Metric Score = 0 + 24 ½ - 26 = -1 ½ Machine 2: Busy until 9:09 and is 11 minutes away  Metric Score = 9 + 11 - 0= 20 Machine 3: Busy until 9:07 and is 5 minutes away  Metric Score = 7 + 5 - 0 = 12 minutes away Machine 4: Available since 8:55 and is 33 minutes away  Metric Score = 0 + 33 - 5 = 28 Machine 5: Busy until 9:09 and is 0 minutes away  Metric Score = 9 + 0 - 0 = 9 Metric Score = Remaining processing time of current job + time to reach customer – time since availability

11. GANTT CHART

12. RESULTS & NEXT STEPS Results:  Citibike required 9 bikes needed for 24 job instances  Our shared car network algorithm required only 6 machines for 24 job instances  No late jobs  We service 100% of all requests Next Steps:  Try out our algorithm with more data (what happens when there are 100, 1000 jobs?)  Play with max lateness and % of requests serviced parameters to see affect on machine requirements  Create a program to compute algorithm

13. QUESTIONS? “General Solutions get you a 50% tip.” Source: xkcd.com