A paper posted on arXiv introduces a mixed reality-based telemanipulation framework that combines human control with semi-autonomous assistance. The goal is to let operators execute fine-grained bimanual tasks with greater ease and precision.
According to the abstract, the approach is designed to be scalable and intuitive, allowing humans to control robots in ways that mirror their own task execution. This is particularly relevant for tasks that require two arms to work together delicately.
Beyond direct control, the system is positioned as a practical way to collect on-robot data for robot learning. By having humans guide robots through fine manipulations, the resulting demonstrations could be used to train future autonomous policies.