Ultrathin neuromorphic soft muscle integrating sensing and disturbance robustness for adaptive object balancing

Ultrathin soft robots enable conformal attachment and lightweight deployment, yet their sensitivity to environmental stimuli can amplify disturbances into persistent oscillations, limiting stability under non-stationary conditions. This paper presents an ultrathin neuromorphic soft graphene muscle (SGM, 106.8 ± 2.79 μm) that integrates self-sensing, actuation, and disturbance rejection within a single device for compact closed-loop control. Proprioceptive states are encoded through real-time graphene resistance changes, while thermally driven deformations provide tunable posture regulation. The SGM further demonstrates practical actuation and sensing-actuation integration through robotic grasping and mimosa-inspired localized tactile responses. Online reinforcement learning using Proximal Policy Optimization enables oscillation suppression and steady-state convergence. Under 2.5 m/s airflow, the SGM reduced fluctuation amplitude by 62.93 ± 16.08%. In a pole-balancing task, it supported 1.19 times its own weight and maintained upright stability for over 13 h, even under sustained 2.0 m/s airflow. On a moving cart at 0.1 m/s and 0.4 m/s, upright stabilization was maintained under base-acceleration disturbances. These results establish a pathway toward deployable ultrathin soft robots operating stably in disturbed environments.