Automation in the food industry has accelerated in recent years, driven by demographic shifts, labor shortages, and heightened safety concerns. While dedicated machines excel at repetitive production tasks such as rice making or chocolate molding, packaging and transfer operations often rely on pick-and-place methods. Industrial robotic arms, valued for precision and speed, are widely deployed, but their high cost and limited adaptability present challenges. The choice and design of end-effectors—the interface between robot and product—remain critical.
Suction cups dominate current factory use due to simplicity and low cost, yet they struggle with moist, porous, or irregular surfaces. This leaves many tasks to human labor, particularly when handling cooked or prepared foods stored in containers rather than aligned on conveyors. Researchers have explored gripping-type end-effectors capable of adapting to diverse shapes and textures. Hygienic design principles, as outlined by EHEDG (2018), demand minimal mechanical complexity to avoid contamination, while maintaining high speed and low cost.
End-effectors can be classified by contact position: top, side, bottom, or combinations thereof. Top-surface grasping includes suction cups, Bernoulli or Coanda effect grippers, adhesion-based devices, and needle penetration tools. Innovations such as jamming grippers (Amend et al., 2016) and soft suction grippers with switchable stiffness (Koivikko et al., 2021) show promise but face commercialization hurdles. Side-surface grasping benefits from better placement accuracy and has seen advances in soft robotics, with commercial examples from Soft Robotics, Inc., OnRobot, and Festo. Specialized designs, such as magnetorheological fluid grippers (Pettersson et al., 2010) and incompressible fluid fingertips (Maruyama et al., 2013), target fragile items.
Bottom-surface grasping is rare in automation due to instability risks, though the SWITL hand from FURUKAWA KIKO demonstrates feasibility for semi-liquid and delicate materials. Hybrid approaches—combining top and side contact—include the RightHand Robotics gripper and Festo’s TentacleGripper, merging suction with mechanical grip. Other combinations, such as side and bottom or full-surface envelopment, aim to stabilize slippery or irregular products, with examples like Applied Robotics’ meat gripper.
Recognition systems are equally vital. When products are aligned on conveyors, 2D image processing and pattern matching suffice, as in ABB’s FlexPicker. For random bin-picking (RBP) scenarios, where items overlap in 3D space, recognition becomes complex. Traditional 3D template matching struggles with deformable geometries, prompting machine learning solutions. Joffe et al. (2019) applied Faster R-CNN to poultry handling, while Nishina and Hasegawa (2020) used deep networks to identify grasp points for irregular items. Low et al. (2021) employed YOLOv3 to detect varied vegetables, achieving notable classification speed.
Fundamental data on food properties underpins effective handling strategies. Parameters such as Young’s modulus, viscoelasticity, friction, and geometry inform grip force, velocity, and end-effector design. Studies have measured elasticity in products from apple pulp to boiled carrots (Ogawa et al., 2015), characterized viscoelasticity in cheese and ham (Singh et al., 2006), and examined friction using tribological systems (Joyner et al., 2014). Geometric modeling supports simulation and recognition, with MRI-based carcass models (Goñi et al., 2008) and shape inspection systems for kernels and crackers (Ding and Gunasekaran, 1994).
Existing food databases focus on nutrition and composition, such as USDA’s FoodData Central, but lack mechanical property data relevant to robotics. Image datasets like UEC FOOD-100 and Food-101, while designed for visual recognition, could be adapted for robotic categorization. A dedicated database of physical and mechanical properties, akin to the Yale-CMU-Berkeley object set for manipulation research, would enable simulation, benchmarking, and end-effector selection.
From a systems perspective, compact, portable designs suit high-mix, low-volume production environments. Ease of use is paramount in factories without dedicated system engineers, favoring intuitive interfaces and minimal maintenance. Integration with Industry 4.0 technologies—IoT and cyber-physical systems—offers pathways to enhanced communication, self-monitoring, and automated updates.
Despite the food industry’s long history, it remains labor-intensive. Progress hinges on combining advances in end-effector design, recognition algorithms, and property databases into cohesive, hygienic, and cost-effective systems. The interplay between mechanical engineering, computer vision, and materials science will determine how quickly robots can match human dexterity in handling the diverse, delicate, and often unpredictable products that define modern food manufacturing.