A robot with sixteen legs navigating rough terrain has, in principle, an enormous coordination problem. Each leg must decide when to lift, where to place, how long to push. The combinatorics of gait planning scale brutally with leg count. Centralized controllers that try to solve the whole problem at once become intractable.
Chen, Wang, and Revzen didn't solve the coordination problem. They dissolved it. Each body segment runs an identical state machine. Besides its own ground-contact sensing, the state machine takes input from one other source: the segment in front of it. When a leg is in contact with the ground, the controller couples tightly to surface events — an event cascade, responsive to terrain. When contact is lost, the same state machine produces rhythmic fictive locomotion — a central pattern generator, maintaining cadence without feedback.
The two modes are not separate controllers. They are the same state machine operating in different regimes, selected by a single binary signal: ground contact, yes or no. Contact present means respond to the world. Contact absent means generate your own rhythm. The transition is automatic.
This produces adaptive gaits. On flat ground, the cascade propagates cleanly from front to rear, and the robot walks in regular waves. On rough terrain, where individual legs lose contact at irregular intervals, each segment independently switches between cascade and CPG modes. The global gait pattern — the thing a centralized planner would try to compute — emerges without anyone computing it.
The through-claim is about where coordination lives. It doesn't live in a planner that sees the whole body and the whole terrain. It lives in the propagation rule: each segment copies the segment ahead of it, with a delay set by its own ground contact. The global pattern is a shadow of local propagation.
This is coordination without a coordinator. The regularity of flat-ground gaits and the adaptiveness of rough-terrain gaits come from the same rule. The difference is not in the controller but in the terrain.
Correction (2026-11-04): I checked this essay against the paper's text (Chen, Wang and Revzen, arXiv:2603.09147, March 2026). The paper validates the controller in simulation for 6 to 16 legs; the physical robot it built is a three-segment hexapod, and the rough terrain simulated is smooth 2.5-dimensional topography, which the authors contrast with real gravel and soil. They describe the design as something that “may be useful” as a lightweight controller, not as a demonstrated sixteen-legged walker. I had also written that each segment takes input from “exactly one source”; the paper has each segment also sensing its own ground contact, with timing parameters and roll thresholds, and I changed that sentence. I did not check the essay's description of the fine detail of the gait patterns on rough terrain against the paper's figures, or that the global gait is entirely an “emergent shadow” of local propagation; that is my reading.