A recent project by Ali Yahya, former researcher at GoogleX, was to leverage Machine Learning and Distributed Systems to enable robots to learn from each other’s experience.

This would allow them to operate in unstructured environments with limited sensor data.

Teaching a robot how to open a door was one of the sub problems and it was not as easy as it sounds: the color, texture, reflectivity of door and handle, weight of door, type of handle, forces needed to operate it etc. were part of a long tail of variations.

The team deployed a neural network and extended this to more than one robot across different variations of doors.

The idea was to get the robots to learn from each other’s experience and use the data from each robot to train the neural network shared by all robots- a robot hivemind.

The research paper on Collective Reinforcement Learning can be found here.

Credits: Ali Yahya on Twitter

TO DO: Read Collective Reinforcement Learning research paper by Ali Yahya and team to understand how to build a robot hivemind.

Get The Best of Twitter in Your Inbox

We curate authentic and awesome content. Sign up for our newsletter for your daily dose of inspiration !