Results and Analysis 2/3 - RL vs. Non-AI Experiment Results
The results for the RL vs. non-AI control experiments can be seen above. The non-AI control is a simple reactive method in which the robot attempts to move directly towards the recharger unless it encounters an obstacle, in which case it takes a random move. As can be seen, RL control outperforms non-AI control in each of the five environments. There is little difference in the overall performance of RL control in each of the five environments and it maintains a higher degree of positive reward relative to non-AI control. In some of the environments the non-AI control performance is very low, or fails completely because the robot cannot reach the recharging station because it becomes stuck in local minima and cannot escape e.g. non-AI E5 and non-AI E4. While non-AI control can produce results that allow the robot to reach the recharging station successfully in some environments, RL control learns a path which is on average shorter. In other words, RL control provides a more efficient as well as more reliable means for directing the mobile robot back to the recharging station.