Humans possess the unique ability to grasp the goals, desires, and beliefs of others, which is crucial for anticipating actions and collaborating effectively. This skill, often known as “theory of mind,” is innate to us but stays a challenge for robots. Nevertheless, if robots are to turn out to be truly collaborative helpers in manufacturing and every day life, they should learn these abilities as well.

In a brand new paper, which was a finalist for the very best paper award on the ACM/IEEE International Conference on Human-Robot Interaction (HRI), computer science researchers from USC Viterbi aim to show robots to predict human preferences in assembly tasks. This can allow robots to in the future assist in various tasks, from constructing satellites to setting a table.

“When working with people, a robot must continually guess what the person will do next,” said lead writer Heramb Nemlekar, a USC computer science PhD student supervised by Stefanos Nikolaidis, an assistant professor of computer science. “For instance, if the robot thinks the person will need a screwdriver to assemble the following part, it will probably get the screwdriver ahead of time in order that the person doesn’t should wait. This manner the robot can assist people finish the assembly much faster.”

A Recent Approach to Predicting Human Actions

Predicting human actions will be difficult, as different people prefer to finish the identical task in various ways. Current techniques require people to exhibit how they would really like to perform the assembly, which will be time-consuming and counterproductive. To deal with this issue, the researchers discovered similarities in how individuals assemble different products and used this data to predict preferences.

As a substitute of requiring individuals to “show” the robot their preferences in a fancy task, the researchers created a small assembly task (known as a “canonical” task) that may very well be quickly and simply performed. The robot would then “watch” the human complete the duty using a camera and utilize machine learning to learn the person’s preference based on their sequence of actions within the canonical task.

In a user study, the researchers’ system was capable of predict human actions with around 82% accuracy. This approach not only saves effort and time but in addition helps construct trust between humans and robots. It may very well be helpful in industrial settings, where employees assemble products on a big scale, in addition to for individuals with disabilities or limited mobility who require assistance in assembling products.

Towards a Way forward for Enhanced Human-Robot Collaboration

The researchers’ goal is just not to exchange human employees but to enhance safety and productivity in human-robot hybrid factories by having robots perform non-value-added or ergonomically difficult tasks. Future research will concentrate on developing a way to robotically design canonical tasks for various kinds of assembly tasks and evaluating the advantages of learning human preferences from short tasks and predicting actions in complex tasks in various contexts, comparable to personal assistance in homes.

“A robot that may quickly learn our preferences can assist us prepare a meal, rearrange furniture, or do house repairs, having a big impact on our every day lives,” said Nikolaidis.