To be successfully deployed in the real world, robots need to have the ability to perform a variety of daily tasks, from housework to industrial processes, including manipulating fabrics, such as folding clothes into wardrobes, or helping elderly people with limited mobility tie before social activities.
, To date, developing robots that can efficiently perform these tasks has been quite challenging.
Many methods used to train robots to complete fabric manipulation tasks rely on imitation learning, a technique that uses videos, motion capture clips and other data from humans completing relevant tasks to train robot controls.
, While some of these techniques have achieved good results, to take it further they often require large amounts of human presentation data.
This data can be expensive and difficult to collect, and existing open source datasets do not always contain as much data as is used to train other computing technologies, such as computer vision or generative artificial intelligence models.
, According to foreign media reports, researchers at Singapore, Shanghai Jiao Tong University and Nanjing University have recently developed new methods to enhance and simplify the training of robot algorithms through human demonstrations.
The researchers outlined this approach in a pre-published paper on the arXiv platform, which allows robots to use the massive amount of videos posted on the Internet every day as human demonstrations of daily tasks.
Photo source: arXiv, return to the first electric network home page>,.
