One simulated weekday of the autonomous shuttle and e-bike fleet, replayed on the streets of our MIT and Kendall service area. Timing, queues, batteries and waits are the simulation's own.
Loading streets and fleet…
08:00
Model day: even arrivals, every ride 1.1 km, only floor(n x availability) vehicles in service (the others sit out), no batteries
Demand
50people per hourLow
08:00
Wait of every passenger
Waited 5 min or lessWaited more than 5 minAverage wait in each 2 h block5 min requirementDemand per hour (relative)Now (click or drag to jump)Rush-hour traffic blocksFaded dots have not happened yet. Vertical axis is square-root scaled
What the fleet is doing vehicles by state, whole day
CarryingDriving emptyChargingIdle (rest of the fleet)
Build your own fleet
Shuttles
E-bikes
Runs the same equations as the Python model in your browser. The random numbers differ, so a day is close to the Python result but not identical.
Fleet
…
Right now
Passengers waiting 0
Longest wait now 0.0 min
Trips done 0
People carried 0
Avg wait so far 0.0 min
Fleet battery 100%
This day vs the model
Event log
Utility of this architecture (SAU and MAU)
Each curve turns one result into a satisfaction score from 0 to 1 (SAU), using the Task 1 curves. The weighted sum of the four scores is the MAU. Markers show the model, the average of 200 simulated days, and the day on screen.
Bikes. Streets only keeps bikes on the same street network as the shuttles. Campus paths lets bikes also use the footpaths, courtyards and crosswalks in the City of Cambridge GIS data (dashed lines on the map). Shuttles stay on the streets, and bikes keep their normal speed. Because the simulation has no geography, the paths enter as a shorter trip: building to building, a bike route is on average 4.4% shorter (0.956 of the street route), measured on the real map. Only fleets with bikes change.
Charging. Where it stands is the original rule: a vehicle charges wherever it finished its last ride, which is not realistic. At the base sends a vehicle with a low battery to the charging base, marked on the map at the Volpe site (Third St, east side of the Kendall parcel, an approximate position you can move). Instead of driving back empty it drives to the base, charges there for the same charging time, and drives back into service, so the trip there and back also keeps it away from passengers. The distance is how far a random street point is from the base (about 1.1 km on average), the same for shuttles and bikes. Go charge below sets the battery level that triggers it (default 20%). The base has as many chargers as it needs, a simplification. Because these runs are simulated live in your browser, a day is statistically the same but not draw for draw identical to the recorded days.
Dispatch. Standard is the original rule: a vehicle driving back empty after a drop-off cannot be reassigned until it arrives. Smart sends that vehicle straight to a waiting passenger (it keeps the distance already driven and skips the rest of the return). Everything else is the same: same demand, same random numbers, same batteries and charging.
How to read it. Each dot in the first chart is one request, placed where it arrived and as high as its wait. The dashed line is the 5 minute requirement. Shaded columns show hourly demand. In the rush-hour day, tinted hours are the traffic hours, when shuttles run at 58% and bikes at 85% of normal speed. The simulation has no street map: it draws random trip lengths and random positions. For this picture each pickup, ride and empty return is placed on real streets with the same length and the same timing, so queues, battery stops and waits match the numbers in the model validation exactly. Streets, buildings and river are open data from the City of Cambridge GIS, and landmark positions are approximate.