Researchers at the University of Colorado Boulder have developed new algorithms aimed at improving the safety and decision-making capabilities of robots working in close proximity to humans. The study, led by Morteza Lahijanian, associate professor in the Ann and H.J. Smead Department of Aerospace Engineering Sciences, was presented at the International Joint Conference on Artificial Intelligence in August 2025.
The research addresses challenges that arise when robots operate in unstructured environments alongside people, such as in manufacturing or healthcare settings. While robots are typically optimized for efficiency in completing specific tasks, human unpredictability introduces uncertainty that can compromise safety. Lahijanian’s team proposes an approach that allows robots to manage this uncertainty by incorporating a concept similar to “regret” into their decision-making processes.
Drawing on game theory, the researchers developed algorithms that allow robots to evaluate the outcomes of different strategies based on the presence and potential behavior of humans. Rather than attempting to guarantee successful task completion, the robots use what the researchers call “admissible strategies,” which prioritize minimizing harm while completing as much of the task as possible. These strategies enable robots to account for a range of human behaviors, from cooperative to error-prone, without requiring precise predictions.
In practice, the robot would assess its own actions in terms of future regret, selecting options that are least likely to lead to undesirable outcomes, particularly those involving human safety. For example, if a robot detects that a nearby human worker is making mistakes, it could attempt to correct the issue without causing harm. If that proves unfeasible, the robot could relocate to continue its task in a safer location.
The research emphasizes the need for robots to adapt to human behavior rather than requiring humans to conform to robotic systems. According to Lahijanian, this flexibility is essential for effective human-robot collaboration across various skill levels and industries.
The findings contribute to ongoing efforts to integrate robotics into sectors experiencing labor shortages or high physical demands, including elder care and industrial labor. The researchers suggest that with appropriate safeguards, such technologies could augment human capabilities rather than replace them.
Photo credit Casey Cass/University of Colorado Boulder
