Hierarchical-POET-DERL

This repo contains implementation of the Hierarchical-POET-DERL algorithm on a custom Biped2dWalker environment.

Paired Open-Ended Trailblazer (POET): Endlessly Generating Increasingly Complex and Diverse Learning Environments and Their Solutions and Embodied Intelligence via Learning and Evolution (DERL) are combined in a hierarchical fashion to co-evolve both environments and agent morphologies.

Optimal behaviour for each agent is learnt using Proximal Policy Optimization Algorithms (PPO).

An article on Uber Engineering Blog describing POET can be found here.

Demo Results

Requirements

• Fiber
• OpenAI Gym
• PyTorch

Run the code locally

To run locally on a multicore machine

./run_poet_local.sh final_test

Run the code on a computer cluster

To containerize and run the code on a computer cluster (e.g., Google Kubernetes Engine on Google Cloud), please refer to Fiber Documentation.