Researchers at Stony Brook University and MIT are making strides in developing robots that can learn and improve their skills over time. This initiative aims to enhance the functionality of home robots by enabling them to adapt and become more intelligent as they assist people.
The research, documented in a paper titled "Embodied Lifelong Learning for Task and Motion Planning," was presented at the seventh Conference on Robotic Learning (CoRL). The study is led by Assistant Professor Jorge Mendez-Mendez from the Department of Electrical and Computer Engineering.
Mendez-Mendez explained, "Part of our solution was to make sure the robot was learning new tasks more effectively." He added that instead of only evaluating a robot's performance after initial training, they focused on how its abilities improved over time.
The goal is to design a system where robots can learn both general skills, such as grabbing objects, and specialized skills, like handling fragile items, while determining which skill to apply in real-time situations.
Further details on this development can be found in an article by Ankita Nagpal on the AI Innovation Institute website.