Computer Vision-Based Object Recognition and Automatic Pneumatic Soft Gripping

During recent years, soft robotic is a new sub-class of the robots. Soft robotic has several engaging features, such as lightweight, low cost, simple fabrication, easy control, etc. Commercial products such as soft grippers are now available to apply in various fields and applications, for example, agriculture, medicine, machinery, etc. This paper proposes a novel method of grasping in soft robotic fields using computer vision to find the shape, size, and angle of the object to define the best type of grasping mode. Random Sample Consensus (RANSAC) was used to iteratively select randomly sampled 3D points to determine the working plane and identify the randomly placed object. Furthermore, we designed and fabricated a 3D-printed pneumatic soft actuator. The ratio of payload over weight is around 16. Experiments showed the proposed computer vision techniques and pneumatic soft gripper are capable of automatically recognize the object shape and perform soft gripping.