Lecture 3 Mobility, Algorithms and Society PDF

Mobility, Algorithms and Society Using technology to enable scalable and sustainable transportation systems Samitha Samaranayake Cornell University Traffic congestion Congestion in the US in 2014:...

Mobility, Algorithms and Society Using technology to enable scalable and sustainable transportation systems Samitha Samaranayake Cornell University Traffic congestion Congestion in the US in 2014: $160 billion in wasted time and fuel (~1% of GDP) 6.9 billion hours of delay 56 billions pounds of additional CO2 emissions (2012) [TTI Urban Mobility Report, 2015] Why does traffic congestion occur? Why does traffic congestion occur? + emperical data Triangular flux Flux vehicle density Why does traffic congestion occur? Traffic control Traffic management centers Coordinated traffic management Smartphone enabled dynamic routing Ramp Metering Local Arterial Traffic Signals Express Lanes HOT/HOV control  Behavioral considerations and challenges The Cartoon Introduction to Economics, Volume1: Macroeconomics Yoram Bauman and Grady Klein Technology in mobility – route planning Driving force – GPS based traffic data Technology in mobility –route planning algorithms Goldberg ‘10 Dijkstra’s algorithm Routing algorithms for very large networks d d s s Source: Goldberg ‘10 Dijkstra’s algorithm (1959) State of the art Speedup factor of ~3 million for continental US sized networks! (70m edges)  What about mass-transit? The forgotten piece in the transportation tech revolution Samitha Samaranayake Cornell University [Image by: Juan Carlos Martinez Mori] [Image from Houston Metro] Ride-hailing Approximately 300,000 shared autonomous cars (vs. ~800,000 passenger vehicles) could satisfy the mobility needs of the entire population, with waiting times within 15-20 minutes at peak hours. [Spieser, Treleaven, et al. RVA’15]  P1: driver waiting for a ride request P2: driver heading to pick up a passenger P3: passenger is in the vehicle [Spieser, Treleaven, et al. RVA’15] August 2019 Ride-hailing systems can increase vehicle miles traveled (VMT) per passenger mile compared to private vehicle ownership  P1: driver waiting for a ride request P2: driver heading to pick up a passenger P3: passenger is in the vehicle [Spieser, Treleaven, et al. RVA’15] August 2019 Ride-hailing systems can increase vehicle miles traveled (VMT) per passenger mile compared to private vehicle ownership Þ Increase in negative externalities (e.g., congestion, emissions, impact on mass transit --> equity)  P1: driver waiting for a ride request P2: driver heading to pick up a passenger P3: passenger is in the vehicle [Spieser, Treleaven, et al. RVA’15] August 2019 What about autonomy? - Advantages: centralized control, safety (fewer accidents), traffic efficiency (e.g. flow smoothing), increased accessibility - Disadvantages: induced demand, potential for increased ride-hailing, urban sprawl How can cities utilize new advances in technology in a more scalable, sustainable and equitable manner? How can cities utilize new advances in technology in a more scalable, sustainable and equitable manner? 1. Enabling and encouraging (high-capacity) ride-sharing How can cities utilize new advances in technology in a more scalable, sustainable and equitable manner? 1. Enabling and encouraging (high-capacity) ride-sharing 2. Improve utilization of mass transit (e.g., first/last mile shuttles) High-capacity ride-sharing systems State: Requests Compatibility: Trip Generation: Trip Assignment: Solution: and Vehicle Status Pairwise Generalized TSP Integer Linear Assignment of Shareability Graph Program (ILP) request to vehicles Subject to system Penalty for constraints unmatched rides High-capacity ride-sharing systems State: Requests Compatibility: Trip Generation: Trip Assignment: Solution: and Vehicle Status Pairwise Generalized TSP Integer Linear Assignment of Shareability Graph Program (ILP) request to vehicles - Requests are aggregated over fixed time intervals (e.g., 5-30 seconds) and the optimization problem is solved in “real-time” - Approach that is followed in industry Subject to system Penalty for constraints unmatched rides Data Driven NYC: Reengineering urban transit // Daniel Ramot and Saar Golde, Via. [YouTube video] (April 2017), https://youtu.be/QkaM27CMep8, accessed Oct. 28, 2020 High-capacity ride-sharing – optimizing at scale State: Requests Compatibility: Trip Generation: Trip Assignment: Solution: and Vehicle Status Pairwise Generalized TSP Integer Linear Assignment of Shareability Graph Program (ILP) request to vehicles - Vehicles may already contain passengers and have different characteristics (e.g., capacities) - Constraints can be user specific Subject to system Penalty for constraints unmatched rides [Santi, Resta et al. PNAS’14] [Alonso-Mora, Samaranayake, Waller, Frazzoli, Rus. PNAS’17] High-capacity ride-sharing - optimizing at scale State: Requests Compatibility: Trip Generation: Trip Assignment: Solution: and Vehicle Status Pairwise Generalized TSP Integer Linear Assignment of Shareability Graph Program (ILP) request to vehicles - Any trip of size k requires a clique of size k in the sharability graph - Requires solving thousands of small instances of a generalized Traveling Salesman Problem (TSP) Subject to system - Penalty Can beforparallelized constraints unmatched rides [Alonso-Mora, Samaranayake, Waller, Frazzoli, Rus. PNAS’17] High-capacity ride-sharing - optimizing at scale State: Requests Compatibility: Trip Generation: Trip Assignment: Solution: and Vehicle Status Pairwise Generalized TSP Integer Linear Assignment of Shareability Graph Program (ILP) request to vehicles Xk ✏i,j Subject to system Penalty for constraints unmatched rides [Alonso-Mora, Samaranayake, Waller, Frazzoli, Rus. PNAS’17] High-capacity ride-sharing - optimizing at scale State: Requests Compatibility: Trip Generation: Trip Assignment: Solution: and Vehicle Status Pairwise Generalized TSP Integer Linear Assignment of Shareability Graph Program (ILP) request to vehicles Xk ✏i,j - ILP formulation does not explicitly model the road network - #variables limited by QoS constraints Subject to system Penalty for constraints unmatched rides [Alonso-Mora, Samaranayake, Waller, Frazzoli, Rus. PNAS’17] High-capacity ride-sharing - optimizing at scale State: Requests Compatibility: Trip Generation: Trip Assignment: Solution: and Vehicle Status Pairwise Generalized TSP Integer Linear Assignment of Shareability Graph Program (ILP) request to vehicles Subject to system Penalty for constraints unmatched rides [Alonso-Mora, Samaranayake, Waller, Frazzoli, Rus. PNAS’17] High-capacity ride-sharing - optimizing at scale State: Requests Compatibility: Trip Generation: Trip Assignment: Solution: and Vehicle Status Pairwise Generalized TSP Integer Linear Assignment of Shareability Graph Program (ILP) request to vehicles Subject to system Penalty for constraints unmatched rides - Idle vehicles are rebalanced to locations with expected future demand [Alonso-Mora, Samaranayake, Waller, Frazzoli, Rus. PNAS’17] High capacity sharing at the scale of NYC in real-time Sample week: - May 5 - 11, 2013 - 380k (Sun) – 460k (Fri) trips/day - 2000 active trips at anytime - Served by 13,580 taxis NYC Network: 4,092 nodes, 9,453 edges A general framework for aggregated movement of people and goods State: Requests Compatibility: Trip Generation: Trip Assignment: Solution: and Vehicle Status Pairwise Generalized TSP Integer Linear Assignment of Shareability Graph Program (ILP) request to vehicles Subject to system Penalty for constraints unmatched rides - Modular system allows easy integration of features (e.g., walk and ride, demand prediction, EV fleets) - Generalizes to related problems (e.g., food delivery, school bus routing, transit applications)  Some observations from pilot deployments - Many pilot deployments of various styles - Small publicly funded short-term deployments - Low occupancy, limited integration with transit infrastructure DOE VTO: Micro-transit/public-transit for coordinated multimodal movement of people - Not enough incentives for avoiding single occupancy trips - Labor costs not proportional to vehicle size - The economics needs to change, e.g., congestion pricing Both innovation (AVs) and regulation (Congestion pricing) can help NSF S&CC: Mobility for all - Harnessing Emerging Transit Solutions for Underserved Communities Demand responsive public transit High-capacity ride-sharing (micro-transit) systems State: Requests Compatibility: Trip Generation: Trip Assignment: Solution: and Vehicle Status Pairwise Generalized TSP Integer Linear Assignment of Shareability Graph Program (ILP) request to vehicles 1 1 2 2 Shareability graph: NYC high-capacity ridepooling. 2 ~450,000 requests with capacity 10 vehicles. ~3000-5000 feasible trips per problem instance. Subject to system Penalty for constraints unmatched rides Systems with more service flexibility State: Requests Compatibility: Trip Generation: Trip Assignment: Solution: and Vehicle Status Pairwise Generalized TSP Integer Linear Assignment of Shareability Graph Program (ILP) request to vehicles 1 1 2 2 Shareability graph: NYC high-capacity ridepooling. Shareability graph: Boston school bus routing. 2 ~450,000 requests with capacity 10 vehicles. 403 students with bus capacity of 72. ~3000-5000 feasible trips per problem instance. ~8.5M feasible trips in problem instance. Subject to system Penalty for constraints unmatched rides - Shareability network is too dense and too many trip configurations are feasible - Need techniques for pruning the search space [Bertsimas, Martin, Jaillet. PNAS’19] [Riley, Legrain, Van Hentenryk. CPAIOR’19] [Guo, Samaranayake. In review ‘20] Naïve pruning of the shareability graph (distance based) - 403 students with 66 bus stops. Bus capacity is 72. - The potential trip lists is ~8.5M. - Runtime ~3 hours. [Guo, Samaranayake. In review ‘20] Naïve pruning of the shareability graph (distance based) - 403 students with 66 bus stops. Bus capacity is 72. - The potential trip lists is ~8.5M. - Runtime ~3 hours. More advanced heuristics - 403 students with 66 bus stops. Bus capacity is 72. - The potential trip lists is ~4.5M. - Runtime ~24 minutes. [Guo, Samaranayake. In review ‘20] School bus routing State: Requests Compatibility: Trip Trip Generation: Generation: Trip Assignment: Solution: and Vehicle Status Pruned Pairwise Pairwise Generalized TSP Integer Linear Assignment of Shareability Graph Program (ILP) request to vehicles 1 1 2 1 2 2 2 Subject to system Penalty for constraints unmatched rides - Framework allows for allocating travelers to multiple modes - There are many offline shuttle routing problems (e.g., work shuttles, paratransit) [Guo, Samaranayake. In review ‘20] Demand responsive public transit State: Requests Compatibility: Trip Trip Generation: Generation: Trip Assignment: Solution: and Vehicle Status Pairwise Generalized TSP Integer Linear Assignment of Shareability Graph ProgramIndustrial (ILP) request to vehicles Residential 1 1 2 1 2 2 Kent Station 2 Subject to system Penalty for constraints unmatched rides - Transit centric online problems also lead to larger problem instances than in ridepooling Demand responsive public transit State: Requests Compatibility: Trip Trip Generation: Generation: Trip TripAssignment: Assignment: Solution: and Vehicle Status Pairwise Generalized TSP Integer Linear Non-myopic Assignment of Shareability Graph assignment Program (ILP) request to vehicles 1 1 1 2 2 1 1 2 2 2 Subject to system Penalty for constraints unmatched rides - The setting we have discussed solves a sequence of online assignment problems that are myopic - Future demand predictions can help system performance - Even more important in settings with higher capacity and low relative demand density [Shah, Lowalekar, Varakantham. AAAI‘20] Fully integrated hybrid transit systems How can cities utilize new advances in technology in a more scalable, sustainable and equitable manner? 1. Enabling and encouraging (high-capacity) ride-sharing 2. Improve utilization of mass transit (e.g., first/last mile shuttles) 3. Designing new multi-modal mass transit systems Multi-modal mass transit (bus) systems Hybrid transit system Operational Efficiency Fixed-line transit Feeder service Sustainability - Ride-hailing/ridepooling Equity - Biking/bikesharing + - Micromobility - Microtransit Multi-modal mass transit (bus) systems Hybrid transit system Operational Efficiency State: Requests Compatibility: Trip Generation: Trip Assignment: Solution: Fixed-line transit and Vehicle Status Pairwise Demand-responsive Generalized TSP Integer Linear transit Assignment of Shareability Graph Program (ILP) request to vehicles Sustainability 1 Equity 2 + 1 2 Subject to system Penalty for constraints unmatched rides  Relatedalgorithms Challenges Efficient research topics in fixed-line for analysistransit and control of networked cyber-physical systems https://humantransit.org/2018/02/basics-the-ridership-coverage-tradeoff.html - Tradeoff between coverage and ridership  Related The Efficient research casealgorithms topics for multi-modal for transit analysis and controlplanning of networked cyber-physical systems https://humantransit.org/2018/02/basics-the-ridership-coverage-tradeoff.html - Demand responsive services can help with: - Time varying demand and uncertainty of demand