Unsupervised Learning of Goal Spaces for Intrinsically Motivated Goal Exploration

TL;DR

This paper introduces a two-stage developmental approach where deep learning is used to learn goal spaces from raw sensor data, enabling more autonomous and effective goal exploration in robots without relying on engineered features.

Contribution

The paper proposes a novel method combining deep representation learning with goal exploration, removing the need for engineered feature spaces and improving autonomous skill acquisition.

Findings

Learned goal spaces match engineered representations in exploration performance

Deep learning enables autonomous goal space discovery from raw sensor data

Approach improves robot skill acquisition in complex environments

Abstract

Intrinsically motivated goal exploration algorithms enable machines to discover repertoires of policies that produce a diversity of effects in complex environments. These exploration algorithms have been shown to allow real world robots to acquire skills such as tool use in high-dimensional continuous state and action spaces. However, they have so far assumed that self-generated goals are sampled in a specifically engineered feature space, limiting their autonomy. In this work, we propose to use deep representation learning algorithms to learn an adequate goal space. This is a developmental 2-stage approach: first, in a perceptual learning stage, deep learning algorithms use passive raw sensor observations of world changes to learn a corresponding latent space; then goal exploration happens in a second stage by sampling goals in this latent space. We present experiments where a simulated robot arm interacts with an object, and we show that exploration algorithms using such learned representations can match the performance obtained using engineered representations.