Marty Milkovits | October 8, 2025 | 5 minute read
At a Glance – Key Takeaways

• Standards and Tools: New modeling standards (e.g., GMNS) are gaining traction, and agencies are experimenting with SaaS products and cloud integrations.
• Calibration: Still a challenge, but continuous approaches, automated methods, and updated guidance could improve trust in results.
• Model Applications: ABMs can add value for project-level analysis when scoped carefully; post-processing methods enhance congestion pricing insights.
• Model Inputs: Agencies are adapting to Census data changes and experimenting with more descriptive input variables.
• AI & Machine Learning: Emerging techniques are automating calibration, reshaping choice models, and accelerating code development.

Detailed Takeaways

Evolving Standards and Software Solutions

• New standards gaining traction: GMNS is being used more widely, often with Jupyter and Colab demos. While helpful, these environments are messy, highlighting the need for a standardized platform.
• Specialized tools integrated into workflows: Products like TREDLite streamline project reviews (though positioning users more as operators), and SimWrapper is being adapted for secure cloud environments.
• Model software improvements: Caliper is developing an embedded population generator and trip generation using Boosted Decision Trees, reducing computation time while mimicking large simulation runs. TransCAD now includes GUI-inspired dashboards for transit systems.

References: GMNS Workshop; Accessibility Analysis Demo (Dr. Andre L. Carrel, Ohio State); TREDLite – Chris Gregerson; SimWrapper – Billy Charlton, Susan Xu

Calibration Challenges and Opportunities

• Calibration remains difficult: None of 19 surveyed agencies described it as “easy,” highlighting ongoing uncertainty.
• Continuous calibration shows promise: Could establish models as a “single source of truth,” though planning timelines make this difficult.
• New tools and data integration: Automated methods and big data sources reduce manual workload and improve transparency.
• Need for updated guidance: The FHWA validation manual is now 15 years old; updated standards would support consistent practice.

References: Model Calibration Panel

Model Applications

• ABMs for project analysis: Effective with careful scoping and validation; sometimes simpler models are more appropriate.
• Congestion pricing and equity: Post-processing with VOT-segmented skims reveals differences in how income groups experience tolling and congestion relief.

References: Enhancing Project-Level ABM Application – Adrita Islam (Fehr & Peers); Congestion Pricing Analysis – Raghu Sidharthan (WSP)

Model Inputs

• Adapting to Census changes: Agencies reallocating block group demographics to TAZs using geocoded addresses preserve accuracy without redesigning models.
• Clarity in variables: Moving from deterministic variables (e.g., age, occupation) to descriptive attributes (e.g., fixed work schedule, caretaker status) could yield new insights for mode choice modeling.

References: Bridging Census Geography and TAZs – Muhammad Salaha Uddin (UTSA)

AI and Machine Learning

• Core modeling techniques are shifting: AI-enhanced choice models, machine learning for calibration (e.g., in UrbanSim), and GPU-based frameworks (Flow Through Tensor) promise faster, more data-driven evaluation.
• AI-assisted coding: GitHub Copilot and similar tools support multi-language coding, documentation integration, and structured code generation.

References: Improving Destination Choice with AI – Vince Bernardin; UrbanSim + ML – Jeffrey Hood; Flow Through Tensor – Xuesong Zhou & Henan Zhu

Check out the conference link!