High-Speed Robotic Sorting of Volumetric Deformable Object deploying a Parallel Manipulator and a Programmable Matrix-like Suction Gripper

Abstract

Volumetric deformable object (VDO) manipulation has been a significant challenge due to the objects' unpredicted shape, texture, and properties. Many industrial applications, particularly in the food processing industry, require precise and gentle handling of volumetric deformable products such as raw meat and fruits. The ability to manipulate VDOs effectively is important for any automation systems in these industries, as it improves food safety, reduces contamination risks, and minimizes waste caused by improper handling. However, modeling VDO is complicated and unsuitable to represent every object. Therefore, this type of object requires gripping strategies that can maintain stability without causing damage. Meanwhile, traditional grippers have difficulty dealing with volumetric soft products due to their unpredictable shape and texture. As a result, a need for a new manipulation solution that can manage those problems.

Addressing the scenario of sorting VDOs like raw meat in the food industry, this research focuses on developing a new robotic system. A programmable matrix-like suction gripper was integrated into a high-speed parallel manipulator to perform long-horizon manipulation tasks such as pick-place and pick-flip. Unlike traditional suction grippers, which apply uniform suction across an entire object, the proposed solution allows for selective activation of suction cups. Along with the hardware, a Vision System and a control system were built to increase the capabilities of the robotic system. A control system was specifically designed to synchronize the robot movement with the activation of suction cups to ensure efficient manipulation. The final results of the system were promising, the proposed suction gripper verified its effectiveness as it was able to grasp objects firmly without damaging them and able to perform complex manipulation tasks. Compared to other traditional grippers, the proposed end-effector has more advantages in handling VDOs.

Furthermore, since VDOs are extremely complex to develop a physical-based simulation, a reinforcement learning algorithm was applied to the control system. By integrating AI-driven systems, the research was able to optimize the suction cup's activation and robot movements and proved the potential for intelligent adaptation in real-world dynamic environments. The result of the approach confirmed its effectiveness, as the picking success rate increased, and the flipping rate showed an improved success trend over normal control methods.

In conclusion, the development of this robotic system contributes to the field of VDO manipulation, particularly in food automation. Future research could further continue the learning process and explore potential applications in other industries such as healthcare and logistics. This study demonstrates that by combining hardware innovation with AI-driven control, robotic systems can overcome traditional limitations and achieve more reliable and efficient manipulation of VDOs in complex real-world environments.