In this project, I implement a machine learning (reinforcement learning) algorithm known as Q-learning to train an agent (in this case a delivery robot) to deliver packages across a warehouse. The project was made for the web browser using React JS and a popular 3D JavaScript library known as three.js.
An interactive demo can be found at https://zmaqutu.github.io/3D-Pathfinding-V...
This project mainly focuses on visualizing the agent's learning process, which includes visualizing information that the agent has learned about its environment at any given moment as well as how quickly it has been able to get that knowledge. This allows us to observe changes in the agent's behaviour as we provide different learning rate or agent curiosity values.
In the video demo below, we train our agent for half a million training episodes (epochs).The delivery robot quickly learns to stay away from the bottom left area with walls and ditches by going around the obstacles through the top right. As we continue training, our agent slowly discovers a "potentially better" path through the middle, although it is less likely to try that route by virtue of how dangerous it is.
Auf dieser Seite können Sie das Online-Video 3D Pathfinding Algorithm Visualizer mit der Dauer stunde minuten sekunde in guter Qualität ansehen, das der Benutzer Zongo Maqutu 17 September 2023 hochgeladen hat, den Link mit Freunden und Bekannten teilen, dieses Video wurde auf Youtube bereits 54 Mal angesehen und es wurde von 1 den Zuschauern gefallen. Viel Spaß beim Betrachtenden Zuschauern gefallen!