"MIT researchers have developed a new algorithm that allows robots to learn an individual’s preferences for specific tasks and adapt accordingly to assist with task completion. This advancement is seen as a step toward creating robots capable of collaborating with humans in everyday tasks.
In current manufacturing plants, humans and robots typically have distinct roles: large robots handle repetitive, heavy tasks within enclosed areas, while humans focus on jobs requiring precision. However, MIT’s Julie Shah, an expert in aeronautics and astronautics, envisions a future where robots and humans work alongside each other, helping each other with tasks to improve efficiency, particularly in areas like airplane manufacturing.
For example, robots could assist by delivering tools and materials to workers, reducing unnecessary walking and idle time. Shah believes robots can handle non-critical tasks, such as material delivery, leaving humans to focus on more complex, skill-dependent work.
Unlike traditional robots that follow rigid instructions, robots working alongside humans must be able to adapt to the worker’s personal style. For instance, in aircraft assembly, different mechanics may have varied preferences for how tasks like sealant application and bolt fastening are performed. To address this, MIT researchers have created an algorithm that allows robots to learn these individual preferences quickly and adjust to assist effectively. The team is currently using simulations to train robots to work with humans and plans to present their findings at the Robotics: Science and Systems Conference in Sydney.
The research team conducted tests using spar assembly — a process of building the main structural element of an airplane wing. The process involves applying sealant and securing bolts, but mechanics may differ in the order of operations. The team’s robot, FRIDA, developed by ABB, can adapt to these preferences, either applying sealant or fastening bolts according to the worker's style.
The researchers used a decision tree model to map out potential choices a mechanic might make during the assembly process. Once the robot understands the worker's preferred sequence, it can adjust its actions to match. For instance, if a mechanic applies sealant to all the holes before fastening bolts, the robot learns to follow this approach.
In real-world settings, the robot could undergo initial training and, once familiar with the worker’s habits, recognize them via RFID tags and customize its assistance accordingly. This ability to adapt could not only enhance manufacturing efficiency but also assist in medical environments, where robots could anticipate a surgeon’s needs during a procedure.
Steve Derby, a robotics expert at Rensselaer Polytechnic Institute, praises the algorithm for bringing human-robot collaboration closer to reality, highlighting its potential to transform industries. Shah believes this research is crucial in creating a seamless, efficient collaboration between humans and robots, with applications extending to various fields, including healthcare.
This work, funded by Boeing Research and Technology and conducted in collaboration with ABB, marks an important step toward achieving true robot-human collaboration."
