Binary Space Partitioning as Intrinsic Reward

TL;DR

This paper introduces a novel unsupervised feature extraction method using binary space partitioning, which computes an intrinsic reward signal for reinforcement learning in humanoid robots by hierarchically organizing sensory features.

Contribution

It proposes a new approach combining binary space partitioning with information gain to enhance autonomous learning in high-dimensional sensory environments.

Findings

Effective reduction of sensory data dimensionality.

Improved reinforcement learning performance.

Hierarchical feature representation enhances learning efficiency.

Abstract

An autonomous agent embodied in a humanoid robot, in order to learn from the overwhelming flow of raw and noisy sensory, has to effectively reduce the high spatial-temporal data dimensionality. In this paper we propose a novel method of unsupervised feature extraction and selection with binary space partitioning, followed by a computation of information gain that is interpreted as intrinsic reward, then applied as immediate-reward signal for the reinforcement-learning. The space partitioning is executed by tiny codelets running on a simulated Turing Machine. The features are represented by concept nodes arranged in a hierarchy, in which those of a lower level become the input vectors of a higher level.