Skip to content

tangledprogramgraphs/capstone-2025

Repository files navigation

Tangled Program Graphs (TPG)

Developer Names:

  • Cyruss Amante
  • Calvyn Siong
  • Mark Cruz
  • Edward Gao
  • Richard Li

Date of project start: September 12, 2024

This project is developing an interface to test the evolutionary machine learning framework Tangle Programming Graphs (TPG) in a robotic simulation engine called MuJoCo created by Google Deepmind.

Key Features

  1. Integration with new complex environments like Mujoco

TPG framework will have numerous integrations between basic agents and several environments provide by Mujoco

  1. CI/CD integration to improve development practises

The TPG framework will have an integrated CI/CD pipeline that will automatically run linting processes, build project to ensure it's compatible on different platforms with new changes and run automated unit tests

  1. Experiments to measure reinforncement learning performance

Series of experiments to evaluate the behavior and performance of agents controlled by the TPG algorithm within the MuJoCo environment.

The folders and files for this project are as follows:

docs - Documentation for the project refs - Reference material used for the project, including papers src - TPG source code cloned using Git Subtree src/src/cpp - Contains different experiments and models (Classic Control and MuJoCo) src/scripts - Contains environment commands containing useful scripts for running and plotting experiments test - Test cases

Background

The code in src reproduces results from the paper:

Stephen Kelly, Tatiana Voegerl, Wolfgang Banzhaf, and Cedric Gondro. Evolving Hierarchical Memory-Prediction Machines in Multi-Task Reinforcement Learning. Genetic Programming and Evolvable Machines, 2021.

To learn more, read this PDF.

About

Evolving Quadruped Robot Controllers with Emergent Tangled Program Graphs

Resources

License

Contributing

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Contributors 7