Mixed reality-enhanced spatial admittance control for human-robot collaboration using preferential Bayesian optimization

Admittance control is a widely used method for regulating interactions between humans and robots. The admittance controller comprises multiple parameters that can be adjusted to improve the overall system performance. In this study, a novel human-centered approach is presented for the adjustment of admittance parameters in human-robot collaboration. First, a head-mounted display with mixed reality features is integrated into the human-robot collaboration system, which enables the user to interactively adjust the mixed reality components using hand gestures in a real-world space. The admittance parameters are then spatially adjusted based on the configured mixed-reality components, which significantly enhances the task configuration and execution efficiency of the human-robot collaboration system. Subsequently, the admittance parameters are fine-tuned using a human preference-based approach. Specifically, preferential Bayesian optimization is used to optimize parameters based on user preferences without directly evaluating or defining the objective function. Pairwise comparisons of parameter sets by humans are used to build a preference model, which helps efficiently identify the optimal parameter set. The proposed approach is user-friendly and accessible to novice users and requires minimal programming knowledge. Experimental validations and demonstrations were conducted using a 7-degree-of-freedom robot arm, and the results showed the effectiveness of the proposed method in enhancing human-robot collaboration.