Facebook is working on AI that will be trained using self-learning methods

tech2 News StaffMay 21, 2019 07:57:26 IST

Facebook is expanding its areas of research in robotics to build better AI with the help of self-supervised learning.

As Engadget reports, the social networking giant shared details about three additional areas of research, including model-based reinforcement learning (RL) to enable robots that can teach themselves through trial and error using direct input from sensors, teaching the ability to explore using human-like curiosity, and how to feel physically using predictive deep-learning model.

Facebook is building an AI that will help robots teach themselves through trial and error and will have human-like curiosity. Representational Image. Photo credit; Reuters.

The researchers were able to get a six-legged robot to teach itself to walk. Using the RL algorithm, the hexapod learnt how to walk in a matter of hours instead of days.

Facebook's AI research lab known as FAIR is pursuing this line of work in robotics technology. The company’s chief AI scientist, Yann LeCun, told Bloomberg in an interview that "FAIR has a duty to see around corners and be prepared for future products and services, including robots."

The FAIR engineers are working alongside computer scientists from New York University. The report states that with the combined help, researchers were able to reduce the number of tries to grab an object from thousands or hundreds of tries to tens of tries.

With inputs from ANI.