Inspiration
We wanted to explore how public datasets, available to everyone, could help identify new possibilities for electric autonomous vehicle fleets. By comparing U.S. metro areas with markets where autonomous services already operate, we set out to build a transparent way to find and explore potential market opportunities.
What it does
ODDyssey combines public data about population, commuting, climate, charging access, and roads to rank candidate markets. Users can inspect the evidence behind each ranking, adjust the scoring weights, and see how their choices affect the results.
For the top-ranked market, users can run a seven-day hypothetical fleet simulation. They can adjust assumptions such as fleet size, demand, fares, and random seed, then explore results including completed rides, wait times, utilization, mileage, revenue, and charging.
How we built it
We collected public data from sources including the U.S. Census, NOAA, the Department of Energy’s Alternative Fuels Data Center, Census road-network files, and a Rhode Island travel-model layer for the Providence simulation scenario.
Offline data adapters validate and combine source records into consistent metro-level profiles. We preserve raw snapshots, record provenance, and keep missing values explicit rather than treating them as zero.
Our ranking model groups 15 measurements into Familiarity, Readiness, and Opportunity. It applies documented transformations, including $\log(1+x)$ for skewed counts and densities, then normalizes values against frozen bounds. The pillar weights are 40%, 20%, and 40%, respectively. These weights are our modeling choices, not a private operator’s methodology.
A FastAPI backend serves the validated data and deterministic ranking and simulation calculations. Pydantic defines the API contracts, and a Next.js dashboard presents the rankings, evidence, explanations, and simulation playback.
Challenges we ran into
Public datasets use different formats, units, and geographic boundaries. We had to harmonize them carefully while keeping each measurement tied to its source. Some desired measurements were unavailable or incomplete, so we kept those gaps visible instead of estimating values without evidence.
We also had to distinguish market screening from operational prediction. The Providence travel-model layer provides a spatial pattern for simulated requests, but request timing and fleet behavior are assumptions. It is not real ride-hail demand or operator telemetry.
Accomplishments that we're proud of
We built an end-to-end workflow from public data to ranked markets and an interactive fleet simulation. Users can trace rankings back to their underlying evidence, change the scoring weights, and reproduce a simulation with the same inputs and seed.
We are also proud that the project makes its limitations visible. A score or simulated outcome is not a safety assessment, deployment approval, or claim about a private operator’s system.
What we learned
We learned that data quality, provenance, and clear assumptions matter as much as the scoring model. Normalizing different kinds of measurements makes comparison possible, but it does not remove uncertainty or make every source equally complete.
We also learned to keep public-data market screening, hypothetical simulation, and real-world deployment decisions distinct.
What's next for ODDyssey
We want to expand source coverage, improve geographic detail, and add more evidence about transit, airports, and other factors that may shape fleet opportunities. We also want to refine simulation profiles and make assumptions easier to inspect and compare.
Any future ranking or simulation improvements will remain grounded in public evidence, reproducible methods, and clearly stated limitations.
Log in or sign up for Devpost to join the conversation.