Muscle coactivation-inspired stiffness-driven control strategy for antagonistic soft robots

Soft robots excel at safe and adaptive interaction but are often constrained by limited load capacity and manipulation precision. Antagonistic soft actuators offer a promising route to address these challenges by coupling actuation with variable stiffness, inspired by biological systems such as elephant trunks and octopus tentacles. However, existing control strategies largely treat stiffness as a passive mechanical property, whereas those that incorporate stiffness regulation typically rely on analytical models, limiting their deployment in soft robots with difficult-to-model dynamics. Here, we introduce a muscle coactivation-inspired, data-driven control strategy that explicitly integrates motion control with stiffness modulation, enabling precise and robust control without reliance on analytical models or system-specific tuning. The robot requires only minutes of learning from simple motions to acquire diverse functionalities, including grasping, stiffness sensing, liquid pumping, and reconfigurable bridging. By explicitly harnessing stiffness regulation, this work establishes a general and scalable control paradigm that expands the functional scope of antagonistic soft robots beyond passive compliance.