Unsupervised Learning and Exploration of Reachable Outcome Space

TL;DR

TAXONS is an unsupervised, task-agnostic exploration algorithm that learns diverse policies in high-dimensional spaces by building a low-dimensional outcome space, improving exploration in sparse reward reinforcement learning.

Contribution

Introduces TAXONS, a novel divergence-search method that explores outcome spaces without task-specific knowledge, using autoencoders to guide policy diversity.

Findings

Successfully finds diverse controllers covering the outcome space

Operates without prior knowledge of the outcome space

Effective in high-dimensional observation settings

Abstract

Performing Reinforcement Learning in sparse rewards settings, with very little prior knowledge, is a challenging problem since there is no signal to properly guide the learning process. In such situations, a good search strategy is fundamental. At the same time, not having to adapt the algorithm to every single problem is very desirable. Here we introduce TAXONS, a Task Agnostic eXploration of Outcome spaces through Novelty and Surprise algorithm. Based on a population-based divergent-search approach, it learns a set of diverse policies directly from high-dimensional observations, without any task-specific information. TAXONS builds a repertoire of policies while training an autoencoder on the high-dimensional observation of the final state of the system to build a low-dimensional outcome space. The learned outcome space, combined with the reconstruction error, is used to drive the search for new policies. Results show that TAXONS can find a diverse set of controllers, covering a good part of the ground-truth outcome space, while having no information about such space.