Ardian Jusufi of the Institute of Neuroinformatics has been selected in the Global Call shortlist for the 2026 Breakthrough of Falling Walls. The international recognition highlights his highly interdisciplinary research at the interface of biomechanics, robotics, and physical AI — revealing how animal bodies contribute directly to intelligent movement, and how these principles could make robots more agile, robust, and safe.
Bodies answer before brains do. The fastest corrections to a slip, a crash landing, or a stiffer-than-expected step aren’t neural — they’re built into muscle, tendon, and cuticle. A gecko crashing head-first into a tree trunk has only fractions of a second to avoid falling; there is not enough time for the brain to calculate every force and prescribe every movement. Instead, the tail presses against the surface like an emergency fifth leg, and muscles, compliant body, and contact with the tree collectively generate a stabilizing response. This interaction between neural control, body mechanics, and the environment lies at the heart of the shortlisted project, “Breaking the Wall of Reflexes and Preflexes.”
Animals solve this problem partly through preflexes: immediate, state-dependent mechanical responses produced by muscles, tendons, compliant tissues, body shape, and contact with the environment, which begin stabilizing the body before delayed neural correction dominates. The principle extends beyond compliant muscles and tails to claws, adhesive pads, spines, and scales — as in recent work on the scaly-tailed flying squirrel, whose directional tail scales engage the substrate and provide an additional support point during arboreal locomotion. This is distributed mechanical feedback: useful responses emerge wherever the body meets the environment.
These biological principles are increasingly relevant to physical AI and robotics. Rather than relying entirely on sensors, processors, and rapid feedback control, robots can be designed so their mechanics already favour useful adaptive responses — letting software and physical structure divide the task according to their respective strengths.