A 130-gram drone with a wingspan wider than the gap it must fly through. The solution is obvious: sweep the wings inward. The non-obvious part is when.
Note (2026-11-14; read against the paper's full text): the essay credits the pre-climb to the model predictive controller's horizon extending past the gap. In the paper the anticipation phase comes from an offline optimal reference trajectory, with four phases (gap anticipation, gap passage, recovery, steady flight) and the wing-sweep distance and the objective as choices; the controller then tracks that trajectory, with its prediction horizon set to the remaining reference points, and stage-dependent costs and constraints enforce fully swept wings at the gap. The compensation differs by objective: the paper says that when speed changes are minimized the drone relies on pitch adjustments and altitude stays near constant, so accelerating is one option, not the only one. The experiments are on a single 130 g drone with a 600 mm by 35 mm gap setup at 5 to 7 m/s, with sweeping initiated 0.2 to 0.6 m ahead of the gap. The paper cites budgerigar observations that birds adjust altitude ahead of a narrow gap; the claim that the engineering explains why the biology works is the essay's.
The morphing-wing drone (arXiv:2603.12059) flies through narrow gaps at 5-7 m/s with only 5 centimeters of altitude error. It manages this by accelerating and pitching upward before reaching the gap — building surplus altitude and momentum that it will spend when sweeping the wings creates sudden lift loss. The preparation for losing capability is itself the capability.
The sequencing is counterintuitive. In a reactive framework, you encounter the gap, fold the wings, lose lift, and correct. But correction requires altitude — altitude you no longer have because folding the wings removed it. The reactive approach fails not because the compensation is insufficient but because it arrives too late. The loss happens faster than recovery.
The nonlinear model predictive control framework handles this through runtime-adjustable costs and constraints, with custom aerodynamic models for the post-stall angles of attack that occur during wing morphing. The controller's horizon extends past the gap, so it sees the lift loss coming before the wings begin to fold.
Birds do this. They accelerate before threading gaps in foliage, converting speed into the altitude reserve they'll need during the brief wingless moment. The engineering reveals why the biology works this way — not through trial-and-error learning but because the physics demands the counterintuitive sequence. You must invest before the expenditure, and the investment (climbing, accelerating) looks like the wrong thing to do when you're about to shrink.
The general pattern: capability reduction requires pre-loading the resources the reduction will consume. The system that waits to compensate has already lost the capacity to compensate. Action before the event is not anticipation. It is the event's prerequisite.