Faster than Real-Time Machine Learning within High Fidelity Simulations

Imagine using a virtual learning environment to remove the programmer from the process of developing code for mechanical movement. Efficient artificial intelligence combined with a high fidelity simulation would allow the computer to discover valid, optimal actions for a robot in faster than real-time, thus eliminating the need for human guess-and-test. This paper presents the challenges of developing such a system, and describes a robotic machine and associated simulation that gives testimony to this possibility.