Cognitive Load Reduction in Commanding Heterogeneous Robotic Teams

Abstract

With the proliferation of multi-robot systems, the interfaces required to operate them have become increasingly complex compared to those used for single robot systems. This can present challenges for operators who need to extract relevant information in order to make informed decisions about how to operate the robots. To address this issue, this thesis explores a variety of strategies aimed at improving the intuitiveness and usability of such systems. These strategies encompass a range of approaches, from designing user interfaces to integrating physical input devices, knowledge representations, and other modalities to assist operators. In this context, the thesis proposes a decision support system that provides operators with additional information in an intuitive way, focusing specifically on handling a set of distinct commands for a heterogeneous robotic team. A key constraint during the development of this system was the lack of historical data available to train the modules on. As a result, the proposed system was tested in a few-shot environment and was specifically designed for this circumstance. The support system comprises two modules: one that probabilistically classifies the next command using a data mining approach called sequence prediction, which is used to reorder the available commands in the interface; and a second that creates higher-level commands by mining frequent sequences from the historical dataset. These command sequences are presented to the operator, who can add them as additional executable commands. To evaluate the advantages and disadvantages of this novel approach, a user study was conducted, which showed that both modules increased the efficiency and usability of the system, while also identifying opportunities for further improvement.