A two-lane road can carry twice the traffic of a one-lane road. A three-lane road, three times. Capacity scales in discrete steps because lanes are discrete: you either have one or you don't. The lane is a coordination mechanism — it tells drivers where to go so they don't need to negotiate. But the mechanism wastes space. The gap between lanes, the buffer zone, the moment when the signal is red and the intersection is empty — all capacity lost to the coordination overhead.
Remove the lanes. Remove the signals. Let autonomous vehicles use the entire street width, coordinating computationally through a multi-agent conflict-based planner (arXiv:2603.19952). OptWULF combines optimization with window-based reservation: each vehicle claims a spatiotemporal window, the planner resolves conflicts, and vehicles weave through the intersection on continuous trajectories rather than fixed paths.
Note (2026-11-15; read against the paper's full text): the paper's higher-capacity-for-the-same-width statement for lane-free traffic is an analytical result for roadways under assumed minimum headways; the intersection results are simulations of vehicle-only traffic against two lane-based benchmarks (a signalized intersection and the lane-based SlotIIC). In those simulations the capacity advantage is shown most clearly by FERSTT, an earlier lane-free method the paper also simulates, and the paper says OptWULF's capacity is slightly lower than FERSTT's (its conflict resolver is relatively simplistic). What it reports as OptWULF's distinguishing results are the even use of the whole street and intersection area and nearly constant capacity under asymmetric demand. 'Weave' does not appear in the paper.
The result: lane-free traffic achieves higher capacity than lane-based traffic for the same street width. Capacity scales continuously with width rather than in discrete steps. The intersection area is utilized evenly — no empty lanes while others queue. Asymmetric demand (more traffic from the east than the west) is handled without capacity loss, because vehicles aren't constrained to symmetric lane assignments.
The structural insight: lanes are a human-legibility constraint, not a traffic-flow optimization. They exist because humans need simple rules to avoid collisions — stay in your lane, stop at red. Remove the human, and the constraint becomes pure overhead. The coordination that lanes provide can be done better by computation, which doesn't need spatial discretization or temporal synchronization via signals. The street becomes a fluid, not a pipeline.