“Normality; show me normality, Señor Caballero, and I will show you an exception to the abnormal order of the universe; show me a normal event and I shall call it miraculous because it is normal.” Those words, written by celebrated Mexican author Carlos Fuentes in his epic novel Terra Nostra, apply to virtually every aspect of human behavior, including everyday movements. 

That inherent unpredictability has been a barrier to safe interactions between humans and robots. Now, scientists at MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL) have developed a robotic algorithm that harnesses unpredictability to improve the safety of human-robot collaboration. The team will present a paper regarding their work virtually at the 2021 Robotics: Science and Systems conference.

Robots don’t sense movement or respond out of instinct; they require careful programming to predict and adjust to moment-to-moment human behavior. For example, a robot assistant designed to assist a human with tasks like eating or dressing could cause injury if a collision occurs because the robot failed to accurately predict the human’s movements. Yet standard safety measures can result in a robotic “freeze” when it senses movements that could result in a collision. 

Unlike many existing robots, which rely on programming that responds to a default model of “normal” human behavior, CSAIL’s innovative algorithm includes many models of human behavior. Based on these models, the program analyzes in-the-moment input from the human. The results of this analysis trigger changes in the robot’s movement trajectory; you could say it has a range of choices that it can choose from to adjust for human unpredictability.

The AI algorithm uses data from each interaction to improve its library of human behavior. As the robot’s range of choices expands, it becomes safer and safer. The scientists incorporated a two-pronged approach when programming those reactions; the robot either chooses to avoid a collision altogether or to lessen the force of impact when the collision is inevitable. That’s not unlike how the human brain reacts in the same situation.

CSAIL’s many robotic research projects, including this origami-like, robotic gripping device that can pack objects into a bag, align well with the new algorithm. These solutions could potentially come together in a robot that significantly benefits people who need help due to age, disability, or limited mobility. The next phase of the project will involve models that incorporate how a person’s mood and emotions affect the way they move during robot-assisted tasks.