Assessing Boundaries of AI Planning Models of Human-Robot Collaborative Riveting Processes in Industry-Like Conditions

Riveting is an essential process for the pre-assembly as well as the final assembly of aircrafts. In many cases, the riveting process fails to be fully automated, for instance, in parts with complex geometries. Thus, manual riveting is still widely common. Several works have been carried towards semi-automatic riveting solutions, namely in riveting the section barrel of the aft section to its pressure bulkhead. In [1], a method of communication-free semi-automated riveting is proposed where a partially autonomous robot performs counter-holding while a human worker rivets. The method has been modeled and tested extensively in simulation. However, although a demonstration prototype has been developed, models have been tested on it curtly. This paper investigates experimental evaluation of different model variations on demonstration. The aim is to identify the boundaries of the method in real world conditions. Experimental setups are tailored similarly to the ones used in simulation for comparability purposes. Multiple riveting tests are carried out to assess the applicability of the solution in an industry-like environment. These tests allow online tracking of the worker-robot dynamic. Additionally, they quantify the impact this solution on both task ergonomics and efficiency in a semi-automated process with respect to a manual one.