26th March 2022 Robot peels banana with deep learning Researchers at the University of Tokyo have developed a new machine learning system for a two-armed robot – enabling it to identify, pick up, and peel a banana. The manipulation of deformable objects, like soft fruits, is problematic for robots, because of difficulties in object modelling and a lack of knowledge about the type of force and dexterity needed. Heecheol Kim, a researcher in the Intelligent Systems and Informatics Laboratory, University of Tokyo, worked with colleagues to develop a system based on "goal-conditioned dual-action deep imitation learning (DIL)". This can learn dexterous manipulation skills using human demonstration data. The task involves nine stages – from grasping the banana to picking it up off the table with one hand, grabbing the tip in the other hand, peeling it, and then moving the banana so the rest of the skin can be removed. The complete process takes less than three minutes. The system is data efficient, according to Kim, because it uses only 13 hours of training data, rather than hundreds or thousands of hours: "It still requires quite a lot of expensive GPUs [graphics processing units], but by using our structure, we can reduce the large amount of computation [required]," he says. In addition to bananas, the system could be used more generally for robots to handle tasks that require fine motor skills, the team explains. A paper by Kim and his team is published this month on the pre-print server arXiv.
Comments »
If you enjoyed this article, please consider sharing it:
|
