The New Database To Help Robots Learn

Databases such as ImageNet have long been the bedrock of the AI revolution we’re experiencing today.  With 14 million or so images, they provide a vast repository of content with which to train algorithms.  It’s a trick that roboticists are attempting to replicate with a new database, known as RoboNet.

The project, which is documented in a recently published paper, contains a host of annotated videos showing robots performing various tasks.  The aim is to create a kind of digital brain that allows robots to instantly learn from the actions of other robots.

The team begin each task by recording how the robot performs it, before then taking numerous additional videos to provide an ample supply of data with which to help other machines master the task.  They aim to overcome the data challenges that have hampered the ability for machines to learn complex tasks, especially when a new work environment often requires learning from scratch.

“The common practice of re-collecting data from scratch for every new environment essentially means re-learning basic knowledge about the world—an unnecessary effort,” the researchers say.

Open to learn

The data, which will be made freely available to all, aims to make robotic education a shared experience.  The project has been kicked off with around 15 million video frames of various tasks being uploaded, but the obvious hope is that many other teams will contribute their own content to the database.

“This work takes the first step towards creating robotic agents that can operate in a wide range of environments and across different hardware,” the researchers explain.

There are undoubtedly various hurdles facing RoboNet that ImageNet didn’t have to overcome, purely due to the differing nature of the environment.  For instance, there’s no real clarity over the best way to train robots today, but the researchers hope that RoboNet will go some way towards inspiring attempts to develop a scalable approach to reinforcement learning so that the complexities of real world environments can be tackled.

Check out the video below to learn more about the project.

Facebooktwitterredditpinterestlinkedinmail