The operating system that couples every individual's intent with the physical world of urban mobility.

Population-level mobility data has existed for decades. Cities know when and where demand concentrates. Platforms know which routes are busiest. This understanding has made transportation marginally more efficient at the aggregate level.

But individual-level mobility intent has never been understood — or even collected. Who you are. Why you are moving at this moment. What you actually need, not just where you are going. No system has ever tried to understand this. The data simply does not exist.

RoamingOS is building the intelligence layer that understands mobility at the level of the individual, and coordinates physical assets accordingly. The system does not simply respond. It learns. Every outcome feeds back into its understanding of each person, making the next prediction sharper, the next arrangement more precise.

47
structured signals captured per mobility interaction.
91%
fleet utilization under our orchestration, against 52% for reactive dispatch.
Predicted
Confirmed
AV carrying

Aoidos

Our first product is Aoidos — a mobility agent that understands what each person needs and reaches into the physical world to deliver it before they have to ask. We chose urban mobility because it is the domain where individual intent must translate into physical action — vehicles dispatched, routes coordinated, infrastructure adapted in real time — to matter at all. Aoidos is the instrument we built to close that loop between intent and the world that moves.

Research

The orchestration layer is built on published work in robust reinforcement learning for urban transit, simulation-based optimization for city-scale transportation networks, and multi-agent coordination for autonomous fleet operations. The individual intelligence layer is new territory — prior literature does not exist on this problem because the data required to study it has never been collected.

Team

Built at the intersection of transportation science, machine intelligence, and systems engineering — with research roots in the JTL Urban Mobility Lab at MIT, ITS PATH and the ML Dynamics Lab at UC Berkeley, and the Urban Freight Lab at UW.

MITUC BerkeleyUniversity of Washington
Publications informing this project
Robust Reinforcement Learning Strategies with Evolving Curriculum for Efficient Bus Operations in Smart Cities
Simulation-based Optimization for Vertiport Location Selection: A Surrogate Model with Machine Learning
Simulating Integration of Urban Air Mobility into Existing Transportation Systems: Survey

We are based in San Francisco.