This article presents a comprehensive framework for executing primal cuts on pigs within a Meat Factory Cell (MFC) context, with potential applications for small and medium-sized producers. The framework begins by creating a 3D model from CT-scans, which is then aligned with a 3D point cloud acquired from an Intel© Realsense™ camera using an initial coarse estimate, and refined through Bayesian Coherent Point Drift. Cutting trajectories are generated based on a custom 3D model of the cutting surface, designed with consideration of the pig's skeletal structure and the cutting properties of the knife tool attached to the robot. A qualitative evaluation of the cuts performed by a professional butcher reveals promising results, while also identifying areas for improvement. The article underscores the potential of integrating CT-scans, 3D point clouds, and cutting models to automate primal cuts in the meat industry, addressing the inherent anatomical variability among animals.

The goal of this work has been the development of a deep-learning model which may be used in conjunction with a novel so-called “Meat Factory Cell” platform, namely to enable an industrial robot within the system to identify and successfully grip the limbs of an entire pig carcass. The model consists of three main components: (1) a U-Net-based deep learning model that predicts heatmaps with a probability distribution of gripping and key point locations on the limbs, within RGB-D images; (2) a post-processing element for the extraction of keypoints from heatmaps, and transferral of these points into 3D space using a pinhole camera model; and (3) gripper orientation estimation, which uses the predicted limb key points to define gripper orientation in 3D space. The proposed system demonstrates high precision and robustness in estimating gripping points on pig limbs based on a data test set, which includes two gripping definitions: Norwegian and Danish. These gripping definitions account for variation in the slaughter process in two different European countries. The Norwegian definition gives mAP(0.5…0.95) = 0.971, mAR(0.5…0.95) = 0.982, and distance error 13 mm, while the Danish definition gives mAP(0.5…0.95) = 0.985, mAR(0.5…0.95) = 0.995, and distance error 14 mm. The model was validated in practice during experimental trials at the Meat Factory Cell test facility at the Norwegian University of Life Science (Ås, Norway), with whole pig carcasses (n=25).

Automation is critically important for sustainability in meat production, where heavy reliance on human labour is a growing challenge. In this work, a novel robotic Meat Factory Cell (MFC) platform presents the opportunity for unconventional automation in pork meat processing, particularly abattoirs. Instead of following line-based approaches, which are the main option today, it uses robotics and Artificial Intelligence (AI) to perform complex cutting and manipulation operations on entire unchilled pork carcasses, with awareness of biological variation and deformation. The long-term goal of the MFC is to take a pork carcass as an input and produce seven primal outputs: hams, shoulders, saddle, belly and entire organ set. However, the MFC platform is under continuous development – therefore, this paper aims to demonstrate it through a specific use-case: shoulder removal. The system is evaluated based on data from testing and development sessions (June–November 2022), with a total of 34 attempted shoulder removals. Data regarding the MFCs’ ability to handle variation, in addition to success rate and process timing models are presented. Qualitative feedback from skilled butchers is also discussed. The authors propose that, as well as technical development of the platform, it is important to consider new ways of comparing unconventional systems with their conventional counterparts. Innovative manufacturing systems have more to offer than raw speed and volume; traits such as flexibility, robustness and scalability – particularly economic scalability – should play a prominent role. Future legislation and standards must also encourage innovation rather than hinder innovative robotics solutions.

Robotic automation in the medical industry is complex because of the biological nature of the processed materials and the risks to the patient. Everyone is different, which means that each individual has their own characteristics; different size, bone structure, joint positions, etc. This makes it hard for robots to autonomously operate on humans. RGB-D consumer devices and computing power are revolutionising the way robots interact with the environment. With the camera’s intrinsic parameters and the depth frame, a point cloud, i.e., a set of points in a Cartesian space $\mathrm{IR}^{3}$, can be generated giving more complete information to the application in real-time. This work investigates a GAN (Generative Adversarial Network) to generate the internal surface of an ex-vivo porcine left ham, as a precursor to consideration of human models. That is important as a prediction of the internal structure. A good internal ham surface could be generated even with a small dataset. The complexity of the shapes in the generated data are shown and structures like the ball-joint attachment can be seen.

The robotization of the pig carcass slaughtering process requires the possibility of automatic identification of the pig’s body position and orientation of its individual parts for further gripping and manipulation of the limbs. This paper presents a method for locating the gripping points on the pig limbs based on pose estimation of a pig carcass fixed in a meat factory cell from RGB-D images of carcasses taken from 6 different views. A deep learning model based on U-Net architecture was proposed to solve the problem of keypoint detection to estimate the pose and gripping points of pig carcasses. The proposed method demonstrates high precision and robustness in estimating the gripping points of pig limbs: Norwegian style gripping points - mAP(0.5…0.95) = 0.9504, mAR(0.5…0.95) = 0.9688, distance error is within 15 mm; Danish style gripping points - mAP(0.5…0.95) = 0.9831, mAR(0.5…0.95) = 0.9937, distance error is within 15 mm.

Advances in visual sensor devices and computing power are revolutionising the interaction of robots with their environment. Cameras that capture depth information along with a common colour image play a significant role. These devices are cheap, small, and fairly precise. The information provided, particularly point clouds, can be generated in a virtual computing environment, providing complete 3D information for applications. However, off-the-shelf cameras often have a limited field of view, both on the horizontal and vertical axis. In larger environments, it is therefore often necessary to combine information from several cameras or positions. To concatenate multiple point clouds and generate the complete environment information, the pose of each camera must be known in the outer scene, i.e., they must reference a common coordinate system. To achieve this, a coordinate system must be defined, and then every device must be positioned according to this coordinate system. For cameras, a calibration can be performed to find its pose in relation to this coordinate system. Several calibration methods have been proposed to solve this challenge, ranging from structured objects such as chessboards to features in the environment. In this study, we investigate how three different pose estimation methods for multi-camera perspectives perform when reconstructing a scene in 3D. We evaluate the usage of a charuco cube, a double-sided charuco board, and a robot’s tool centre point (TCP) position in a real usage case, where precision is a key point for the system. We define a methodology to identify the points in the 3D space and measure the root-mean-square error (RMSE) based on the Euclidean distance of the actual point to a generated ground-truth point. The reconstruction carried out using the robot’s TCP position produced the best result, followed by the charuco cuboid; the double-sided angled charuco board exhibited the worst performance.

This paper presents a pig carcass cutting dataset, captured from a bespoke frame structure with 6 Intel® RealSense™ Depth Camera D415 cameras attached, and later recorded from a single camera attached to a robotic arm cycling through the positions previously defined by the frame structure. The data is composed of bags files recorded from the Intel's SDK, which includes RGB-D data and camera intrinsic parameters for each sensor. In addition, ten JSON files with the transformation matrix for each camera in relation to the left/front camera in the structure are provided, five JSON files for the data recorded with the bespoke frame and five JSON files for the data captured with the robotic arm.

The Meat Factory Cell (MFC) concept restructures the slaughter line into cell stations and merges elements of the slaughter- and primal cutting processes. With the MFC approach, most of the primals are removed prior to evisceration. This study describes the effect of the MFC concept on carcass hygiene, carcass yield, meat quality traits, and sensory characteristics of selected MFC products from trials with the very first pig carcasses processed with the MFC approach. Results show that hygiene of MFC carcasses rivals conventionally slaughtered carcasses. For quality variables and sensory characteristics of selected MFC products, the study shows that the MFC approach will result in products that equal, and in some cases surpass, conventional products, provided that proper processing, packaging and chilling is applied.

This paper presents work relating to the reconstruction of 3D point clouds taken from cameras placed arbitrarily in pairs on either side of an object (a pig carcass in this specific use-case). This poses a challenge for complete object reconstruction since the two sides or halves of the object have little or no overlap and high degree of symmetry. The investigation was part of a data collection inside slaughterhouses for a meat factory cell concept development, where the production line cannot be stopped to perform calibration of the cameras. In addition, the footprint should be small, and the cameras' lenses protected and easy to clean. This work presents the method, or pipeline, used to overcome this challenge along with the performance of the pipeline by comparing real-world and point cloud distance measurements.

This paper provides a brief overview of the novel Meat Factory Cell and discusses its concept in the context of increasing sustainability in the meat sector. Job quality, environment, health risks, industrial development and education are discussed as sustainability goals that can be mapped against some of the United Nations Sustainable Development Goals (SDG). Technology can arguably help to improve related processes on a societal level, and to achieve the SDGs.

Background: Meat has been an important protein source for human nu-trition for thousands of years and will continue to be. According to the Organisation for Economic Cooperation and Development and Food and Agriculture Organisation's (OECD-FAO)outlook report 2018-2027, the meat consumption increased around 20% in the last ten years, and it is expected to grow another 15% for the next ten years. The harsh working environment in abattoirs and meat factories, such as cold and wet operating rooms and long difficult handling of heavy loads, contributing to the shortage of a skilled labour forces. This, coupled with a considerable increase in the meat consumption, paved the way for novel approaches in the meat industry to ad-dress this challenge, with robotisation and automation of the meat factories being a necessary change. Scope and approach: In this work, we review the current state of robotisa-tion for the meat industry and its adaptability to a new production concept called the meat factory cell (MFC). The reviewed systems are: (a) Frontmatec AiRA Robots for pork slaugh-terlines (b) Mayekawa Hamdas-RX for deboning pork ham; (c) SCOTT Automated Boning Room for lamb slaughterlines; (d) SRDViand (Systemes Robotis'es de D'ecoupe de Viande) Z-cut robotic system to do the separation of the hindquarter and the forequarter; (e) SRDViand ham deboning system; (f) SRDViand ECHORD-DEXDEB, a robotic butcher left hand; (g) SINTEF GRIBBOT, a chicken fillet harvesting robot. Key findings and conclusions: The slaughterhouse processes can be highly automated, with products available at the market, while the meat processing plants are mainly manual; deboning and fine cuts require a higher level of dexterity comparing to primary slaughter cuts, this leads to more researches of intelligent systems. In the other hand, a new way to process the slaughter products can lead to innovation and smarter systems on slaughterhouses, with the benefits to open new opportunities for smaller producers.

This paper presents the novel Meat Factory Cell (MFC) concept which is being developed in both semi- and fully-automated forms. The MFC provides several important opportunities for the red meat sector, including enhanced robustness, scalability and flexibility. Moreover, it is mindful of the need for small-medium meat processors requiring access to automation, which has proven uneconomical until now. The industry has renewed interest in such automation initiatives, particularly considering its need to improve resilience in the face of future global pandemics. The paper describes the progress of the MFC, as well as a rudimentary framework for realising the implementation. Finally, the paper discusses some of the major hurdles faced in the future.

In the last decade, there has been a rising interest in cyber physical systems (CPS), both in civil and military use. Together with the popularization and evolution of navigation sensors, as global position system (GPS) and inertial measurement unit (IMU), has triggered researches in control systems that are more suitable to non-linear systems of an airplane in flight. Classic control systems has been used from simplified mathematical model. As dynamic systems may present complex behaviour, new concepts and technologies of advanced control systems needs to be used. In this work is shown a control method that combine the reasoning of fuzzy logic with the learning of neural network. An adaptive neuro-fuzzy inference system is described to control the heading of an unmanned aerial vehicle (UAV) with a delta type fixed wing with reduced control surfaces called elevens. The UAV has an embedded autopilot board with a micro-controller (mcu) capable of real-time data processing and navigation sensors making this platform suitable for autonomous flight.