“In robot manipulation, learning is a promising alternative to traditional design methods and has shown great success, especially in pick-and-place tasks,” says Cui, whose The work focused on the interface of robot manipulation and machine learning. “While many research questions remain to be answered, learned robotic manipulation could potentially bring robotic manipulators into our homes and businesses. Maybe in the near future we’ll see robots mopping our tables or organizing cupboards. “
In a review article in Science robotics called “On the way to learned robot manipulation of the next generation”, Cui and Trinkle Summarize, compare and contrast research on learned robotic manipulation through the lens of adaptability, and outline promising research directions for the future.
Cui and Trinkle emphasize the usefulness of modularity in learning design and point out the need for appropriate representations for manipulation tasks. You also find that the modularity allows for customization.
According to Cui, those in traditional engineering may doubt the reliability of learned skills for robotic manipulation as they are usually black box solutions, meaning researchers may not know when and why a learned skill will fail.
“As our article points out, an appropriate modularization of the manipulation skills learned can open up” black boxes “and make them more explainable,” says Cui.
The nine areas that Cui and Trinkle suggest as particularly promising for improving the capacity and adaptability of learned robot manipulation are: 1) Representative learning with more sensing modalities such as tactile, acoustic and temperature signals. 2) Advanced simulators for manipulation to be as fast and realistic as possible. 3) Task / skill adjustment. 4) “Portable” task displays. 5) Informed exploration for manipulation, where active learning methods can efficiently find new skills using contact information. 6) Continuous exploration or an opportunity for a learned skill to continuously improve after using the robot. 7) Massively distributed / parallel active learning. 8) Hardware innovations that simplify more sophisticated manipulations, such as B. skillful manipulations in hand. 9) Real-time performance as manipulation skills learned are ultimately tested in the real world.
Following some of these instructions, Cui and Trinkle are currently working on tactile sensorimotor skills to make robotic manipulators more dexterous and robust.
For Cui, one of the most exciting discoveries he made while exploring current research is that learned robotic manipulation is still in its infancy.
“That gives the research community plenty of opportunities to explore and benefit from,” says Cui. “The promising future and the large space for exploration will make the manipulation of learned robots an exciting research area in the coming decades.”
To learn more about their work, read: Advanced robot gripping, skillful manipulation, and soft robotics.
Comments are closed.