The Optimal Audio Interface for Teleoperation on an Autonomous Farm

Abstract

As robots become more prevalent, designing an efficient communication system for humanrobot interaction becomes an important yet challenging problem. Visual and tactile interfaces are very common in autonomous robots and intelligent systems; however, audio-based interfaces are a relatively new and developing area. We study the scenario in which a fleet of agricultural robots need to communicate a failure case for a human operator to diagnose and respond to in a teleoperation setting. These robots must have a simple yet effective communication system so farmers that may not have robotic experience can operate them. In this thesis project, we develop an agbot simulation platform and various audio communication techniques and characterize the most effective and natural interface. First, autonomous farms of varying complexity are created using the OpenAI Gridworld simulation. Then, a user study with 11 participants is conducted with this simulation to test three audio communication methods: sounds, single-word commands, and full natural language communication. As the robots on the farm experience and report errors, the human is tasked with diagnosing them and keeping the robots going. Afterwards, the user completes a survey to determine the overall effectiveness of the system. The results suggest that the human’s perception of the system is mainly impacted by the audio communication technique not the complexity, and the single word commands provide the best interface. However, not all the results were statistically significant, potentially because of the small sample size, and further studies should be conducted on this topic to confirm the results