Two robots must install 352 screws into a 2.4m × 6m timber slab. The adhesive curing time is fixed — if a screw isn't driven within the window, the bond fails. The robots share the same workspace. They cannot occupy the same space at the same time.
The scheduling framework (arXiv:2603.20577) solves this by organizing tasks into levels — spatiotemporally disjoint sets where no two tasks in the same level can produce a collision. Robots execute all tasks within a level in parallel, then synchronize at a barrier before advancing to the next level. Collision avoidance isn't computed during execution; it's guaranteed by construction of the level structure.
Note (2026-11-12; read against the paper's full text): the paper says that a single robot is often infeasible at this scale because the number of screws can violate the adhesive time windows, which is why it uses two robots; the essay's closing statement that speed is not the bottleneck and coordination is goes against that reading, and I found no statement in the paper that the robots are already faster than humans or that the construction industry needs the method. For the scheduling comparison, the paper reports that the monolithic constraint-programming formulation failed to find a solution within an 1800 second limit for instances of more than 300 tasks, and that its method scales steadily; the word combinatorial does not appear. The paper's framework uses two specialized algorithms (an iterative temporal relaxation and a bi-level decomposition), a 10 percent safety buffer on the time window, and says the barriers give a safe state for human intervention after a task failure.
The result is 108 subroutines completed under tight adhesive deadlines. The computational scaling is steady with problem size, unlike monolithic scheduling approaches that grow combinatorially.
The key insight is the barrier mechanism. Rather than continuously monitoring both robots' positions and replanning when conflicts arise, the system pre-computes which tasks are safe to execute simultaneously. The barrier is a synchronization point — both robots must finish their current level before either enters the next. This trades real-time flexibility for construction-time safety.
It's the scheduling equivalent of traffic signals vs. roundabouts. Signals (barriers) introduce waiting time but guarantee no collisions by temporal separation. Roundabouts (continuous monitoring) are more efficient on average but require constant negotiation. When the cost of collision is a failed adhesive bond and a ruined structural element, the signal wins.
The construction industry needs this. Timber prefabrication is growing, adhesive windows are non-negotiable, and the robots are already faster than humans. The bottleneck isn't speed — it's coordination.