Multithreaded Asynchronous Deep Reinforcement Learning With Multisensor Fusion for Robot Collision Avoidance.

Summary: Imagine a robot trying to walk through a busy room without bumping into anything. Scientists have created a smart new brain for robots to help them do just that! They used a special computer learning method that lets the robot practice in a virtual world before trying it in real life. The robot uses lasers and maps to "see" its surroundings and guess if it might crash. By practicing in both simple 2D and realistic 3D computer worlds, the robot learned how to dodge moving objects safely. When tested in the real world, the robot was great at avoiding crashes while still reaching its goal!