July 31, 2024
Artificial intelligence (AI) research supervised by Associate Professor Daniel Szafir was evaluated at the Mars Desert Research Station in Utah. The study involved an autonomous robot navigating the Mars-like terrain to assess its own abilities and communicating results to a human user assigning its tasks.
The research forms part of Nicholas Conlon’s doctoral dissertation in the Department of Computer Science at the University of Colorado Boulder. Szafir is a dissertation adviser, alongside Nisar Ahmed, an associate professor in the Smead Aerospace Engineering Sciences Department at CU Boulder. The algorithmic techniques being tested enable an autonomous robot to assess its own proficiency and communicate the results to humans in order to help humans understand the robot’s capabilities and the actions it takes. Given initial data about its environment and a task to complete, a robot can relay an assessment of its ability to complete the task back to the human user. As the robot undertakes the task and gathers new or contradictory data about the environment, it can update its assessment, which may cause the user to adjust the task or continue without adjustment. The goal is for a human user to clearly grasp how competent the robot is at the given task.
The Mars Desert Research Station (MDRS) leverages the desert of southern Utah to simulate the surface of Mars, hosting several multi-week missions each year. During MDRS missions, the astronauts wear space suit simulators anytime they leave the habitat structure (HAB), use only navigation technology that would be available in a mission on Mars, and experience realistic delays when communicating with personnel who would be based on Earth.
During the two-week evaluation period in late April, MDRS crew #297 was provided a Clearpath Jackal robot and a laptop to use in science gathering activities outside the HAB. Each activity involved the robot navigating to a site and collecting data from the site for approximately one hour. Astronauts could use data from the robot’s self-assessments to monitor progress, adjust goals, and abort activities if needed.
The study collected data related to how the communication of the robot’s own competency assessments affected the human user’s situational awareness of the robot. The study will help researchers in the field of human-robot interaction to better understand why and how human users rely on robots during critical science missions in harsh environments. The results will guide development of future robotic systems by informing how they can be made more explainable and trustworthy, as well as how they can be tailored for maximum utility in space missions.