Fleet Rebalancing 2D — Demand-Predictive Shuttle Repositioning

A grid of 36 city zones cycles through morning and evening demand peaks, spawning ride requests as a Poisson process weighted by each zone's live demand. A small fleet of autonomous shuttles serves those requests — but the way idle vehicles spend their downtime is the whole experiment: in Reactive mode a shuttle simply parks wherever its last trip ended and waits to be summoned; in Predictive mode a greedy demand-weighted assignment recomputed every few seconds drives idle shuttles toward the highest-scoring under-covered zone before the next surge of requests lands. Watch the live average wait time, coverage percentage and cumulative repositioning distance to see the trade-off between proactive driving cost and passenger wait — the same trade-off transit-network operators tune when they decide how aggressively to rebalance an on-demand shuttle or micromobility fleet. This top-down 2D rebuild also fixes an accounting bug in the original 3D version's repositioning-distance counter (see "How it works" below) and adds drag-to-pan / scroll-to-zoom over the zone grid.