2026-02-24 Australia Team

[joecastiglione]

Agenda

• Phase 11C
  • Proposal Review Process
  • Final Task Orders
• ActivitySim Testing Plan
  • ActivitySim Testing Plan Issue
  • Engineering Team Discussion Notes
• SANDAG Presentation
  • SANDAG Presentation
  • Product / Community Team Meeting Notes
• Product / Community Team Membership

Notes

Attendees: Joe (Zephyr), Shaun Tabone (TfNSW), Amir (TfNSW), Rathish Vijayarajan (TfNSW), Paris (Victoria), Kate (Victoria), Joel Freedman (RSG), others

1. Phase 11C Task Orders

Task orders for Phase 11C were finalised and sent to consultants on the day of the meeting. The process involved soliciting competitive proposals across seven tasks, followed by two weeks for review, two weeks for consolidation of feedback, and a final week to draft and review task orders. Key outcomes:

• ActivitySim Application & Analysis Guide – Awarded to CS, who proposed a conservative approach plus one additional network change scenario.
• EET Work (two tasks) – Assigned to both Outer Loop and RSG, whose proposals were seen as complementary. Work covers a more efficient random number generator and its implementation (the latter as an optional task), investigation of alternative sampling strategies, and other performance enhancements
• Automated Calibration – Awarded to RSG, praised for a robust and compelling design that builds systematically on existing work. RSG tasks also include contributions from Jeff Newman (Driftless) and WSP.

3–5 people reviewed each set of task proposals.

2. TfNSW / Victoria EET Investigation Update

Victoria (via BLC) engaged to implement the RSG EET into their model and compare outputs against a baseline. Key findings:

• Upgrading from ActivitySim v1.3 to v1.5 and enabling explicit chunking (with support from Rathish) delivered simultaneous reductions in memory requirements (512 GB → 128 GB machines) and improvements in runtime — an unexpected win-win.
• Upon implementing the EET, runtime increased by approximately 50× (versus the expected 2–3×), concentrated in the location choice pre-sampling step.
• The suspected cause is the CT-RAMP shadow pricing method, which is computationally intensive due to continuous iterative sampling.
• Tim (VLC) proposed a Poisson sampling approach: rather than sampling zones sequentially, it calculates the probability of each zone being included in the sample and draws all zones simultaneously, greatly reducing computational load.
• This work is investigative and scoped only to the location choice pre-sampling component. Victoria is committed to keeping the consortium informed and aligning any changes with the core codebase. Tim has been invited to brief the broader consortium.

3. Testing Plan

A GitHub issue outlining an initial testing plan scope has been created by Jeff Newman. Key discussion points:

• Current testing relies heavily on full-system CI tests rather than targeted unit tests, which is slow and limits development agility.
• Testing is acknowledged as a complex and unresolved issue across the project.
• The shared Google Doc (linked from the agenda) is open for comment. Members are encouraged to review Jeff's outline and the engineering team meeting notes and add their questions.
• This is expected to be a standing agenda item for the next several engineering meetings as Phase 12 ramps up.

4. SANDAG Presentation Recap

SANDAG presented at the previous Product & Community Team meeting, notable as the first US MPO to use ActivitySim in an adopted Regional Transportation Plan. Highlights included ancillary models developed by RSG: an international border crossing demand model, an updated commercial vehicle model (including last-mile deliveries), and an airport ground access model. Documentation links were circulated by Bargav post-meeting. Members are encouraged to review the recording and presentation.

5. Consortium Membership Update

Mark Moran (MWCOG / Washington D.C. MPO) is retiring and will be replaced on the Product & Community Team by his colleague Feng.

6. Growing the Consortium

Shaun raised the question of the strategy for expanding consortium membership. Key points from Joel:

• Several US agencies (including SERPA/South Florida, Kansas City MPO, and Oregon Metro/Portland) are implementing or have implemented ActivitySim without contributing to the consortium. Joe has started to engage these agencies and their consultants directly.
• Outreach is also underway in Europe, where activity-based modelling is less mature. Presentations have been made to the European Association for Activity-Based Modeling; a presentation to the UK Modellers Transport Forum is planned for late April; an abstract for a September conference on ABMs and economic appraisal has been submitted.
• TfNSW confirmed they will continue to encourage Queensland's Department of Transport and Main Roads to join.
• Members were invited to flag other target organizations and to suggest strategies. Joe offered to report back on membership growth quarterly.

7. Other Business

• TfNSW submitted three abstracts to the Transport Planning Professionals conference (Brisbane, September). If accepted, they intend to share presentations with the consortium.
• Victoria noted Queensland is using the Victorian model as a donor model with strong initial results, and is working with Outer Loop on a vehicle allocation model — a potential future benefit for Victoria.
• Victoria is planning an industry briefing (within the next couple of months) where consultants will present on model development work.

Actions

#	Action	Owner
1	Share links to US activity-based modelling guidelines and resources (including ABM Primer)	Joe (Zephyr)
2	Update agenda with link to SANDAG meeting recording	Joe (Zephyr)
3	Arrange Tim (VLC) briefing to consortium on Poisson sampling approach	Joe / Victoria
4	Continue to encourage Queensland Dept of Transport & Main Roads to join the consortium	TfNSW
5	Report back on consortium membership growth strategy quarterly	Joe (Zephyr)