
The Atlas robot from Boston Dynamics is constantly being improved and perfected. — Photography Courtesy of Boston Dynamics/AFP Relaxnews
The humanoid robot Atlas never ceases to amaze. Already capable of carrying and moving heavy loads without outside help, it will now undergo special training to enable it to operate completely autonomously by interacting with its surroundings.
Boston Dynamics has joined forces with the Robotics & AI Institute to advance its humanoid robots through reinforcement learning. In robotics, this is a machine learning technique that involves making decisions by interacting directly with the environment. With each action performed, the robot receives a reward or penalty, which helps it to adjust its strategy in order to maximise the rewards gained over time. This type of learning is therefore based on a whole process of trial and error, all without direct human supervision.