| Summary: | This study presents a comprehensive framework for evaluating the quality of human manipulation datasets in contributing to efficient human-robot collaboration (HRC) in manufacturing settings. Recognizing the importance of safety and efficiency, we want to achieve the development of robotic systems capable of planning safe trajectories in anticipation of human actions. We assert that the quality of human motion prediction is contingent upon the quality of the underlying datasets, and thus our work proposes methodologies for appraising human manipulation video datasets based on their adequacy for human landmark extraction, which ultimately impacts their transferability to subsequent analytical models. We examine several intricate assembly operations, then extract relevant human skeletal data and implement a spatiotemporal analysis for effective sample validation and scenario categorization. From this, we extract out the duration taken to complete assembly tasks in an attempt to predict the next task completion duration, allowing for better HRC.
|
|---|