COLLABORATIVE CUTTING: AGILE ROBOT LEARNERS FOR MULTIPURPOSE MEAT INDUSTRY AUTOMATION

The meat packing industry --- like most food processing industries --- is faced with a major challenge. On the one hand, working conditions are often considered suboptimal. On the other hand, the workers must be highly skilled and maintain rigorous job standards: poor worker performance can lead to consequences for animal welfare and food safety.One possible solution isleveraging automated technologies (e.g., robots) as a replacement for or complement to human workers. Automation has been touted as a way to improve worker satisfaction, worker safety, product quality, product uniformity, and other related labor challenges for the meat processing industry. But despite nearly three decades of discussion, there is fairly limited uptake of robotic equipment within American meat and hoofstock facilities. This limited uptake is partially because current technologies are expensive, only do one job, have high infrastructure requirements, and effectively replace a human laborer while doing little to change the working conditions for other people on the processing line. The factors limiting the success of automation are derivative of the historical approach to robotics development: traditionally, robots were constructed to replace humans rather than collaborate alongside teams. But instead of direct replacements for human workers, we envision robots as partners that leverage learning and collaboration to create human-robot teams which are agile, effective, economical, and productive.The long-term goal of this proposal is to catalyze a shift in robotics for meat processing: from single-use human replacements to dynamic and adaptive teammates. More specifically, our goal is to enable robot arms to learn from and communicate with their human coworkers in order to improve the productivity, safety, and standards of food processing. Our central hypothesis is that by using multimodal sensing to capture the behavior of human experts we can rapidly transfer this expertise to robot learners while keeping the human in-the-loop, allowing for an agile and adaptive paradigm that incorporates robots into human teams. We formulated this hypothesis based on pilot studies in which participants used multimodal interfaces to teach commercial robot arms to perform precise manipulation tasks. Our project rationale is that by addressing the way in which robots and humans collaborate on specific meat processing tasks we will more broadly reconceptualize the use, adaptability, and role of robotic tools in food processing. Until we address the basic assumption about how robots can be used within food processing industries and demonstrate the value of a collaborative approach, we will never be able to realize the potential of robotic technologies to improve job quality, satisfaction, safety, animal welfare, product quality, and food safety.More specifically, in this project we propose three key objectives.First, we will develop a multimodal SmartCarve station that uses wearable gloves, instrumented tools, and nearby cameras to collect information about how human experts complete meat processing tasks. Based on our preliminary design and testing, we hypothesize that multimodal combinations of visual sensing and 9 degree of freedom inertial motion sensing can be leveraged to map human motions for effective translation to robotic colleagues.Second, we will develop algorithms that transfer human expertise in meat processing tasks to robot learners. We compare traditional approaches (where the robot replaces a single human) to our proposed collaborative learning method (where the robot learns multiple tasks alongside the human). Based on our pilot studies, we hypothesize that combining learning with control enables an agile and adaptive human-robot partnership while guaranteeing safety and precision.Third, we will create interfaces and systems that robot feedback to human teammates. In our first two objectives we transfer knowledge from the human to the robot. In this last objective we close the loop so that the robot teammate can ask for help and make suggestions to improve the workflow of the human-robot team. Based on our prior work with haptic and visual alerts, we hypothesize that robot feedback will accelerate human teaching and optimize team workflow.