ActivitySim

ActivitySim is travel-forecasting software built by a consortium of US regional planning agencies.^1,2 An activity-based model takes a list of households and the people in them and decides what each person does in a day. It chooses whether they go to work or school, what other outings they make, in what order, how they travel and when.⁶ The outings are grouped into tours, and each tour is broken into trips.^5,6

A consortium of transport agencies leads the project, sets its priorities and provides the resources for development.¹ Most are metropolitan planning organisations, the bodies that plan transport for a US city region.¹ The aim, in the consortium's words, is software "for distribution at no charge to the public."¹ Member agencies that have published their own implementations include the Puget Sound Regional Commission (Seattle), the Atlanta Regional Commission, the Metropolitan Council (Minneapolis-St. Paul) and the Oregon Modeling Statewide Collaborative.² These are not necessarily complete or calibrated tools.² Version 1.6.0 was released in September 2026 and the code is under active development.^1,7

A model needs three kinds of input. The first is the households and persons to simulate, created by a population synthesis program such as PopulationSim.⁶ The second is land use by zone, such as population and jobs.⁵ The third is skims, tables of travel time and cost between every pair of zones by mode and time of day.^4,5 ActivitySim is one part of a larger system. Other software builds the networks and skims and assigns the resulting trips to roads and transit, which ActivitySim does not do.³ The behaviour is set in configuration files of expressions and coefficients, which an agency can re-estimate in the software's estimation mode.^3,5 The key output is a single HDF5 file holding each data table.⁵

It is heavy. The main worked example, based on the San Francisco Bay Area's travel model, has 2.7 million households, 7.5 million people and 1,475 zones, and takes between an hour and a day to run depending on the machine.⁴ One agency's guidance asks for at least 8 cores and 128 GB of memory and recommends 12 or more cores and 256 GB.⁴ Sampling a subset of households makes test runs faster.⁴

For a model of a place, ActivitySim answers who travels where and when if a road closes, a transit line opens or land use changes, at the level of individual people rather than averages. It is a planning tool for agencies with modelling staff, not something to run casually. For simulating every traveller on a road network in detail, see MATSim. The code is BSD-3-Clause.¹