Learning to play Doodle Jump through interaction

A collaborative UC Davis project applying Q-learning to an HTML5 game.

  • Q-learning
  • HTML5
  • JavaScript
Machine Learning & Doodle Jump
Role
University project · Contributor
Period
2015
View source

Overview

I contributed to a UC Davis team project applying Q-learning to HTML5 Doodle Jump. The browser interface includes training-speed controls, performance charts, and saved learning state.

My contribution

Contributed to the collaborative university project exploring how a Q-learning agent learns to play Doodle Jump through interaction.

Outcome

A collaborative browser-based reinforcement-learning project with visible training controls, performance charts, and persistent learning state.

Doodle Jump project report charts showing score per life and states explored during training
Training plots from the university project’s report: score per life and explored states. These are historical results, not a current benchmark.

Q-learning in a game environment

The team applied Q-learning to an HTML5 implementation of Doodle Jump. The browser interface lets users adjust game speed and disable rendering while the agent trains.

Inspecting and preserving training

Charts show average score and explored states per ten deaths. The agent’s learning state is automatically saved and loaded between browser sessions.

The project scope

  • HTML5 Doodle Jump

  • Q-learning agent

  • Learning through interaction

All projects