A tendon-driven robotic jellyfish uses discrete constraints to reach 150-degree bending and reinforcement learning for closed-loop depth regulation

A tendon-driven soft actuator with discrete constraints is proposed, where a single actuator bends up to 150 degrees and shows an approximately linear relationship between tendon displacement and bending. Prior jellyfish-inspired soft robots often trade large deformation against repeatable actuation; this work places both in one actuator by combining a flexible substrate with discrete constraints. The summary reports two structural figures, the 150-degree bending limit and the approximately linear displacement-bending relation, as design-level quantitative descriptions, without fatigue, repeatability, or batch-consistency data.

Eight actuators driven by four servos let the robot perform stable swimming, attitude adjustment, and self-righting. The work extends the actuator-level linear behavior to whole-robot behavior, using few servo channels to drive multiple actuators and achieve several underwater actions rather than propulsion alone. The summary states swimming, attitude adjustment, and self-righting as a behavior list, without quantitative speed, attitude angle, or self-righting time.

A reinforcement-learning controller built on the linear actuation achieves closed-loop depth regulation in both simulation and physical experiments. Many jellyfish-inspired robots remain open-loop or simulation-only; this work uses the linear actuator model as a control foundation and carries closed-loop depth control into physical experiments. The summary explicitly names simulation and physical experiments as the two validation settings, but reports no depth error, convergence time, or trial counts.

The results indicate that mechanical constraints can improve the controllability of soft actuation while preserving compliant jellyfish-like motion, offering a route toward more manoeuvrable and autonomous jellyfish robots. The work reframes constraints from a limitation into a source of controllability, offering a design idea that trades structure for control in soft underwater robotics. This is the authors' overall judgment based on the described design and experiments, a directional conclusion, and the summary provides no comparison against other platforms.

The results are aimed at researchers in soft underwater robot design and control: for those interested in turning soft actuators into modelable, closed-loop drive units through mechanical constraints, or in using reinforcement learning for underwater depth holding, this work lays out a chain from actuator structure to whole-robot behavior to controller. Its intended setting is depth regulation and attitude actions under laboratory simulation and physical experiment conditions, not long-duration sea trials or complex current environments. Readers can use it to judge whether constrained soft actuation is worth adopting in their own platforms.

Because only the summary was read here, figures and experimental details are missing, and open questions remain: how well the 150-degree bending and approximately linear relation hold over repeated cycles, how the depth loop behaves under disturbances or different depth setpoints, the simulation-to-physical gap, and the controller's sensitivity to actuator parameter changes. These are directions for further verification rather than criticisms of the work.