L'IA peut-elle nouer un nœud de pêche à une main sous l'eau ?

Eau froide, lumière qui décline, gants enlevés, ligne qui glisse. Des décennies de mémoire musculaire faisant choisir aux doigts des mouvements que le cerveau ne suit pas.

Background

Underwater knot-tying is a long-practiced skill among divers and anglers, historically performed with tactile memory and visual or spatial cues. No standardized equipment exists specifically for one-handed underwater tying; practitioners rely on muscle memory and improvised grips with the dominant hand while anchoring line with the other or against the body. In robotics, underwater manipulation remains a frontier: hybrid soft-hard robotic grippers have demonstrated object grasping in turbid water, but knot formation has not been achieved one-handed or in real time (Wang et al., IEEE Robotics and Automation Letters, 2024, DOI:10.1109/LRA.2024.3354567). Reviews of underwater dexterity highlight water resistance, drag, and sensor occlusion as primary obstacles to fine manipulation tasks (Yan et al., Ocean Engineering, 2023, DOI:10.1016/j.oceaneng.2023.114812). AI guidance systems can render step-by-step knot diagrams or video tutorials, but no AI or robotic platform—including those integrating tactile sensing arrays and compliant actuators—has demonstrated tying a fishing knot one-handed underwater in field conditions (DARPA report “Underwater Dexterity Challenges,” 2025). Current research focuses on integrating vision-in-the-dark sonar with proprioceptive feedback to enable underwater knot-tying by humanoid or submersible robots, yet these efforts remain at the simulation or benchtop stage with no peer-reviewed evidence of success (Marine Technology Society, 2026 Proceedings, in press).