In fact, the process may vary from one individual to the next. One person may apply sealant in one hole, hammer in a bolt, and repeat the process for each hole. Another may apply sealant to all the holes before starting to hammer in the bolts. One robot that might be programmed to help with this process is ABB's two-armed FRIDA. Its arms have a wide range of motion, which Shah said can be manipulated for fastening the bolts or applying sealant to the holes.
The team developed a decision tree computational model. Each branch represents a choice a mechanic might make. For example, after applying sealant to one hole, does the worker apply sealant to the next hole or hammer a bolt into the first hole?
The team trained a laboratory robot to observe an individual's chain of preferences, learning that person's preferred order of tasks. The robot could adapt quickly and either apply sealant or fasten a bolt according to that individual's style of work.
Shah said that in real-life manufacturing settings, many workers wear radio-frequency identification (RFID) tags. Factory robots can be programmed to recognize people they have worked with before (through RFID tags or another method) and initialize the appropriate task plan for that person.
The group will present its findings in July in Sydney at the Robotics: Science and Systems Conference. The research was conducted in collaboration with ABB and supported in part by Boeing Research and Technology.