Autonomous Mobility-on-Demand (AMoD) simulation and analysis pipeline for Toronto.
Driverless, autonomous vehicles (AVs) pose sweeping implications for the future of our cities and roads, from congestion to the design and usage of public spaces. In Ontario, the Ministry of Transportation has allowed AVs to be tested on public roads under certain conditions. Though it may be years still before fully autonomous vehicles roam our streets, we are at a critical window of opportunity to anticipate and plan for their impacts on our cities and citizens. In this project, I test different operating scenarios of autonomous, mobility-on-demand (AMoD) vehicle fleets in the City of Toronto. This pipeline includes generating a synthetic trips dataset and using the AMoDeus library to create, run, and view agent-based models of AV fleets. After running the simulation, I explore how metrics such as passenger wait times and fleet travel distances vary across different operating policies and income and age groups.
To use data_utils:
cd data_utils
pip install -r requirements.txt
You will need GDAL for the postprocessing notebook. Install by going to
https://pypi.org/project/GDAL/ or running conda install GDAL. To configure AMoDeus, follow the README.md in
amod and
amodtaxi.
Contains code to run the scenario creation and simulation pipeline. Uses amodtaxi to create the initial scenario.
- Create a folder to store the simulation results.
- Set up run configurations, setting the working directory to the created folder above. See amod for more instructions.
Contains code and configuration for creating the MATSim/AMoDeus scenario for Toronto.
To create your own scenario, at minimum you require:
- population (eg. taxi trips dataset)
- location specs consisting of the bounding box, EPSG code, and reference frame center for the area of interest
The scenario settings are stored in the src/main/resources/ folder. The
rest is a data pipeline that runs these steps: loading the map and trip dataset,
building a MATSim network, and creating a population.
Contains the preprocessing code for synthesizing a population/trip dataset and
postprocessing code for cleaning and plotting the results. The input folder
contains the trip synthesis code and all required datasets. The output
folder contains the AMoDeus simulation results, analysis notebook, and plots.
- Global Euclidean Bipartite Matching: Ruch, C., Hörl, S. and Frazzoli, E., 2015. Amodeus, a simulation-based testbed for autonomous mobility-on-demand systems. In 2018 21st International Conference on Intelligent Transportation Systems (ITSC) (pp. 3639-3644). IEEE.
- Demand-Supply Balancing: Maciejewski, M. and Bischoff, J., 2018. Large-scale microscopic simulation of taxi services. Procedia Computer Science, 52, pp.358-364.
- Model-Free Adaptive Repositioning: Ruch, C., Gachter, J., Hakenberg, J. and Frazzoli, E., 2020. The+ 1 Method: Model-Free Adaptive Repositioning Policies for Robotic Multi-Agent Systems. IEEE Transactions on Network Science and Engineering.
Baseline results show that the choice of AV fleet policy does influence how wait times are distributed across income groups. In particular, passengers from lower income households tend to experience longer wait times when a fleet rebalancing algorithm is used. Non-rebalancing policies don't show this tendency, suggesting that the rebalancing policy could potentially exacerbate mobility inequities between passengers in low and high income households. There doesn't appear to be a trend in wait times or travel distances with respect to passenger age groups.
The below results are with a fleet size of 200.
| Wait times across household income | Distances across household income |
|---|---|
 |
 |
| Wait times across age groups | Distances across age groups |
|---|---|
 |
 |
The below results are with a fleet size of 200.
| Wait times across household income | Distances across household income |
|---|---|
 |
 |
| Wait times across age groups | Distances across age groups |
|---|---|
 |
 |