ABM3 Intra-Household AV Allocation and Taxi/TNC Routing Models

[dhensle]

Overview
This PR implements both the intra-household AV allocation and taxi/TNC routing models for the ABM3 system.

Implementation Details

• Intra-household AV allocation model
• Taxi/TNC routing algorithm with optimized vehicle assignment
• Configurable parameters for vehicle occupancy, wait times, and batch processing
• Performance optimizations for large households/trips

Testing Results & Model Performance
The TNC routing model has been tested and validated with the following performance metrics:

TNC Trip Statistics

• Total TNC Vehicle Trips: 413,714
• Total TNC Trips: 205,277
• Total Vehicles: 7,648
• TNC Trips to Vehicles Ratio: 26.84
• Average Trips per Vehicle: 54.09

Service Quality Metrics

• Average Initial Wait Time: 2.60 minutes
• Average Detour Time per Passenger (pooled trips): 4.28 minutes

Vehicle Efficiency Metrics

• Percentage of Deadhead Vehicle Trips: 50.31%
• Average Occupancy (all vehicle trips): 0.95
• Average Occupancy (weighted by VMT): 1.62
• Total Refuel Trips: 3,160 (0.76% of all vehicle trips)

VMT Analysis

• Total VMT from TNC Vehicle Trips: 1,708,092.80 miles
• Total Deadhead VMT: 235,714.59 miles
• Percentage of Deadhead VMT: 13.80%

For runtime improvements, here is a quick comparison from Iteration 3 (100% sample). In total, the

• Java AV allocation model + TNC routing model: 0:36:45
• Python Taxi/TNC routing model + AV and TNC matrix builder: 0:02:30

This translates to a total runtime savings of roughly 90–100 minutes for the AV/TNC routing components across all three iterations.

Below shows the summary comparing between new Python algorithm and old Java algorithm:

The new Python algorithm shows significantly improved fleet efficiency but with trade-offs in service quality:
Key Improvements ✓

• 57% reduction in total vehicles needed (7,648 vs 17,635)
• 147% increase in average trips per vehicle (54.09 vs 21.89)
• 13% increase in trips-to-vehicles ratio (26.84 vs 23.76)

Performance Trade-offs ⚠️

• 27% increase in deadhead trips (50.31% vs 39.45%)
• 13% decrease in average occupancy (0.95 vs 1.09)
• 7% increase in total vehicle trips (413,714 vs 386,084)

The new python algorithm does not have significant impact to mode share and regional VMT, as shown in the figure below from Model Comparison Report:

The new Python algorithm also fixes the bug in Java program (below was from RSG developing work):

[JiaXu1024]

The AV allocation model is not directly integrated with the main ActivitySim model pipeline. Could you update the config file to include it in the workflow?

[Copilot]

Pull request overview

This pull request implements two new routing models for ActivitySim's ABM3 demand modeling framework: an intra-household autonomous vehicle (AV) allocation model and a taxi/TNC fleet routing model. These models simulate vehicle dispatching and repositioning decisions for both household-owned AVs and for-hire vehicles.

Changes:

• Adds AV routing model to allocate household AVs to driving trips and reposition vehicles between trips
• Implements taxi/TNC routing model with trip pooling, vehicle dispatching, and refueling logic
• Creates matrix builder utility to aggregate vehicle trips into OMX matrices for traffic assignment

Reviewed changes

Copilot reviewed 21 out of 24 changed files in this pull request and generated 6 comments.

Show a summary per file

File	Description
src/asim/extensions/av_routing.py	Core AV routing logic for matching household AVs to trips and repositioning decisions
src/asim/extensions/init.py	Registers new av_routing extension module
src/asim/scripts/taxi_tnc_routing/taxi_tnc_routing.py	Main taxi/TNC fleet simulation with pooling and dispatching
src/asim/scripts/taxi_tnc_routing/tnc_av_matrix_builder.py	Utility to aggregate vehicle trips into OMX matrices by time period and occupancy
src/asim/scripts/taxi_tnc_routing/taxi_tnc_routing_settings.yaml	Configuration file for taxi/TNC routing parameters
src/asim/scripts/taxi_tnc_routing/create_test_taxi_tnc_routing_scenario.py	Test data generator for taxi/TNC routing model
src/asim/configs/resident/settings.yaml	Adds av_routing model to pipeline and av_vehicle_trips to output tables
src/asim/configs/resident/settings_mp.yaml	Adds av_routing to multiprocessing pipeline configuration
src/asim/configs/resident/av_routing.yaml	Configuration for AV routing model specifications and coefficients
src/asim/configs/resident/av_trip_matching.csv	Utility specification for AV trip matching model
src/asim/configs/resident/av_trip_matching_coefficients.csv	Coefficients for AV trip matching utilities
src/asim/configs/resident/av_trip_matching_preprocessor.csv	Preprocessor for AV trip matching variables
src/asim/configs/resident/av_repositioning.csv	Utility specification for AV repositioning model
src/asim/configs/resident/av_repositioning_coefficients.csv	Coefficients for AV repositioning utilities
src/asim/configs/resident/av_repositioning_preprocessor.csv	Preprocessor for AV repositioning variables
src/asim/configs/resident/parking_location_choice_annotate_trips_preprocessor.csv	Updates parking eligibility to exclude AV-serviced trips
src/asim/configs/common/constants.yaml	Adds AV routing constants (max duration, parking cost)
docs/design/demand/av-routing.md	Documentation for AV routing model design and configuration
docs/design/demand/taxi-tnc-routing.md	Documentation for taxi/TNC routing model design and outputs
docs/design/demand/resident.md	Updates resident model documentation to reference AV routing
mkdocs.yml	Adds new documentation pages to navigation

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[dhensle]

Were the demand matrix files produced using the updated model as well? i.e. TNCVehicleTrips_[time_period].omx and emptyAVTrips.omx

Yes, these are being created by the tnc_av_matrix_builder.py file. (It was not yet committed when you made this comment, but is there now!)

[dhensle]

@JiaXu1024 I believe I have addressed all of the comments here. We are doing a final integration test and will of course fix any issues that arise from that, but I believe this is functionally complete!