A fully stretchable octopus‐inspired soft robotic tentacle empowered by multimodal tactile sensory toward embodied intelligence

Abstract An octopus possesses the capabilities of superior adaptability with complex environment, dexterous movements, sensitive perception, and distributed neural control, due to the entirely soft body and numerous ganglia in both its brain and tentacles, therefore regarded as an example of embodied intelligence. Here, we present a fully stretchable sensory soft robotic tentacle with suction and grasping abilities. The tentacle is equipped with carbon nanotube‐based suction cup units, and each unit is integrated with triboelectric and strain sensors. Additionally, by employing multi‐granularity scanning deep cascade forest (gcForest) algorithm, through the integration and training of multimodal data, the average recognition rate can achieve 100% for distinguishing different types of fruits, and 98.40% for differentiating objects with varying shapes and/or hardness. In the complex spatial reconstruction task of a checkerboard pattern composed of 9 distinct materials and 25 pattern conditions, the sensory system attains a 97.92% reconstruction accuracy with assist of the gcForest algorithm. Notably, only 40 datasets per category are required for training. Our study highlights the sensitivity and robustness of this octopus‐inspired soft robotic sensory system, superior accuracy of cross‐modal data, and excellence of gcForest algorithm in few‐shot learning, with great promise for human‐robot interaction applications.