RLR 14 - Simulation vs Real-world Training

This session highlights the training possibilities of RL. One is to simulate the whole system- agent, environment and rewarding system, and make the machine to learn. It's easy, fast, cheaper and safer. But, at the end, the RL system must operate in real world situations. When the trained simulation model is applied with the real world data, due to many problems discussed here, the RL may fail to work as expected. On the other hand, training RL in real world environment is relatively harder, slow, expensive and can be dangerous. The video also deals with different types of training.