Triboelectric nanogenerators (TENGs) serve as a key technology in flexible electronic devices, especially tactile sensors, enabling flexible robotic manipulators to effectively interact with their external environment. Persistent challenges include inadequate system flexibility, a limited dataset for recognition, and less-than-optimal recognition accuracy. Herein, an intelligent soft robotic system incorporating triboelectric sensors to accurately acquire object-grasping information is presented. The liquid metal triboelectric sensor, which utilized a liquid metal fingerprint electrode, exhibited excellent performance retention under real working conditions, both in bending and stretching states, and can identify four types of sensory information. Concurrently, a rotary triboelectric sensor featuring a rack-and-pinion structure is employed to augment the measurement of the dimensions and morphology of the grasped object. This innovative multi-sensor fusion design enabled the entire system to acquire a greater amount of sensing information using fewer channels compared to conventional manipulators. Finally, by integrating convolutional neural networks with deep learning technology, the system achieved a target recognition accuracy of 96.67% across 15 distinct objects. This easy-manufacturing and efficient self-powered sensing system with a soft manipulator, endowing it with intelligent sensing capabilities and demonstrating great potential for digital real-time recognition.