Each of these demonstrations is created through RL-Applet, which is built upon RL-Glue and RL-Library. Source code and installation instructions for RL-Applet can be found here. If you have Java implementations of RL-Glue compatible agents or RL-Viz-compatible task environments and would like to see them up here, please do let me know.


Museum curated by Brad Knox

(The Java applet code used here is buggy with certain Linux installations. Sorry about that. If it really bothers you, feel free to download the source and get to the bottom of it.)

In the world of artificial intelligence, an agent is a computer program that senses its environment and takes actions to affect its environment favorably. Examples of agents are mobile robots, chess-playing programs, and even muscle-stimulating prosthetics for people with motor-neural diseases.

A reinforcement learning agent improves the quality of its actions by learning from experience. An RL agent adjusts its behavior based on a digital feedback signal, called reward, that it receives after each action.

On this website, you can watch RL agents learn to play simple games and tasks. Below are links to each demonstration, with the name of the game followed by the type of RL agent.

  1. Acrobatic robot with Sarsa(λ)

  2. Balancing cart pole with Sarsa(λ)

  3. Kermitbot maze with with a random-acting agent

  4. Mountain car with Sarsa(λ)

  5. Tetris with a random-acting agent