In the field of large scale robotic, which is often remotely operated, having direct interaction can be disruptive for applications such as moving heavy loads or 3D printing. A Cable-Driven Parallel Robot (CDPR) is used here in physical Human–Robot Interactions (pHRI) with an admittance-based control strategy to physically interact with a user in tele-operation or in co-manipulation mode. A user experiment involving participants is designed to assess the performance of the human–robot team in a given task completion. Task performance and interaction quality metrics are defined and recorded during experiments with different robot configurations. The novelty is to provide a methodology to compare the configurations based on the performance metrics. The methodology accounts for variations of the metrics along time of use and assert a training effect leading to a progression or a regression of the performances. The experiment apparatus includes a CDPR, a user task composed of targets to reach with the robot and a handle equipped with a force sensor acting as a control input of a fixed admittance control strategy of the robot. Collected data show that the task performances and the interaction quality vary during the experiments and denote different variation profiles among the user population. Distribution of these profiles among configurations are analysed to determine the configuration that has the best training effect on users.