Safely and Autonomously Cutting Meat with a Collaborative Robot Arm

TL;DR

This work tackles labor shortages in meat processing by introducing a flexible, collaborative robot-arm system capable of autonomously or with-human-help meat cutting tasks (slicing, trimming, cubing). Safety is addressed with a bounded workspace and an instrumented knife that detects contacts, while vision-driven planning and a real-time controller enable precise cuts on pork loins. Instrumented-knife results show promising contact detection but reveal generalizability gaps across meat pieces and tasks, guiding future data collection. The vision and control framework demonstrates industry-mandated product dimensions and acceptable expert reception, suggesting collaborative robots can augment meat-processing work without fully replacing human labor. Overall, the study advances safety and practical feasibility for multi-purpose collaborative robots in meat processing, with clear paths for enhancing generalization and safety integration.

Abstract

Labor shortages in the United States are impacting a number of industries including the meat processing sector. Collaborative technologies that work alongside humans while increasing production abilities may support the industry by enhancing automation and improving job quality. However, existing automation technologies used in the meat industry have limited collaboration potential, low flexibility, and high cost. The objective of this work was to explore the use of a robot arm to collaboratively work alongside a human and complete tasks performed in a meat processing facility. Toward this objective, we demonstrated proof-of-concept approaches to ensure human safety while exploring the capacity of the robot arm to perform example meat processing tasks. In support of human safety, we developed a knife instrumentation system to detect when the cutting implement comes into contact with meat within the collaborative space. To demonstrate the capability of the system to flexibly conduct a variety of basic meat processing tasks, we developed vision and control protocols to execute slicing, trimming, and cubing of pork loins. We also collected a subjective evaluation of the actions from experts within the U.S. meat processing industry. On average the experts rated the robot's performance as adequate. Moreover, the experts generally preferred the cuts performed in collaboration with a human worker to cuts completed autonomously, highlighting the benefits of robotic technologies that assist human workers rather than replace them. Video demonstrations of our proposed framework can be found here: https://youtu.be/56mdHjjYMVc

Figures (9)

• Figure 1: A multi-purpose robot arm for meat processing. (Left) A human collaborator places meat in front of the robot. Using an attached camera the robot detects the location of the meat. Under our proposed framework the robot can either plan and execute desired cuts that process the meat autonomously, or the robot can collaborate with the human to determine which cuts to make. (Right) In our experiments we leverage this framework to process a pork loin by slicing it into multiple cuts, trimming the fat from these cuts, and finally cubing the meat.
• Figure 2: Left: safety precautions for avoiding human-robot collision. Here we constrained the robot's motion into a safe operating region above the cutting board. We also designed an instrumented knife for detecting unexpected contacts between the knife and another object. We conducted two experiments using the knife to cut butter and meat. We discuss the safety framework in Section \ref{['sec:M1']} and the results of the experiments in Section \ref{['sec:RS']}. Middle: testing the error between the robot's planned motion and the desired cut (i.e., the robot's cut precision). This is discussed in Section \ref{['sec:M2']}. Right: example meat processing using our vision and control framework. We enable the robot to autonomously detect meat and fat (vision) and then control its motion to cut the meat into various products. We test our approach on four meat cutting operations: slicing, removing fat autonomously (trimming) or in collaboration with humans (point-to-point), and cubing. More details of our vision and control framework are provided in Section \ref{['sec:M2']}. The results of the meat processing with this framework are listed in Section \ref{['sec:RT']}.
• Figure 3: Schematic showing the robot's safety framework. The robot cuts the meat placed on a cutting board in front of it. (Left) We define a safe operating region around the cutting board, shown here in red, in which the robot is restricted throughout the operation. Staying within this bounded region helps to avoid undesirable contact with humans. (Right) We designed an instrumented knife for detecting undesirable contacts of the knife. This can inform the robot to re-plan its trajectory whenever needed to avoid collision with humans. To validate the effectiveness of the knife we perform two experiments: first, a human cuts butter with this knife; second, a human cuts meat with this knife. We test how accurately our setup can predict contact.
• Figure 4: (Top) Our proposed robot vision and control framework. A camera is mounted on the robot arm. The gripper of the robot holds a knife for cutting. During calibration we use the position of the four markers in the camera frame and the robot frame to optimize Equation (\ref{['eq:M1']}) and find the optimal parameters $\theta^*$. (Middle) Robot motion in three dimensions. The output of the vision module is a trajectory in $x$-$y$ coordinates. While following this trajectory, the robot pauses at intervals and moves in the $z$ direction according to a sinusoid of time period $T$. This up-and-down motion moves the robot in and out of the meat to perform cuts. (Bottom) Example meat processing tasks. We design experiments with three tasks that are representative of meat processing applications in industry. First, meat is sliced into thin strips. Then, any excess fat on the strips is cut off; this is done through interaction with human coworkers (point-to-point) or completely autonomously (trimming). Finally, we produce uniform cubes from the slices (cubing).
• Figure 5: Results from our experiments testing vision and control accuracy on the robot arm. (Left) The three images show example trajectories for the three motions: vertical, horizontal, trimming. The trajectories planned by the vision module are shown in gray and the trajectories executed by the robot arm are highlighted in purple. (Right) Error between the desired trajectory and the robot's actual motion. The gray region shows the industry allowable threshold on Error (khodabandehloo2012roboticsnollet2006advanced).