7 April 2026 in SSRN Electronic Journal
Joint handling of rigid objects by humans and collaborative robots requires the robot to dynamically adapt its trajectory to accommodate variable human behaviour. In this study, human motion is modelled using a probabilistically data-driven method trained on experimental datasets, and it is represented in a low-dimensional latent space that captures task-consistent motion variability without explicit role labelling. Then, the corresponding full human motion is reconstructed in the Cartesian Space, and the human hand trajectory is extracted. During execution of the human-robot collaboration task, the robot follows a generated hand trajectory and continuously evaluates deviations arising from physical interaction based on force, velocity, and spatial consistency. Sustained deviations between the robot's movement and the human co-worker are interpreted as changes in dynamic interaction and drive online trajectory adaptation via a trajectory-joining strategy. Such sustained deviations are interpreted as a transition in interaction role, reflecting a change in motion initiative within the physically coupled system and triggering online trajectory adaptation via a latent-space trajectory-joining strategy. This paper presents a role-transition-aware trajectory adaptation approach for human–robot joint object handling, motivated by manufacturing and industrial collaboration scenarios. Laboratory-based experimental validation on a collaborative robot performing joint object-handling tasks demonstrates bounded trajectory deviations, autonomous recovery toward task-consistent motion after human intervention, and reduced interaction forces compared to a free-drive baseline. These results indicate that trajectory-level latent motion representations can be a potential alternative for ergonomically acceptable human–robot collaboration in industrially relevant joint handling tasks.