Goal-Aware Cross-Entropy for Multi-Target Reinforcement Learning

TL;DR

This paper introduces a goal-aware cross-entropy loss and goal-discriminative attention networks to improve multi-target reinforcement learning by enhancing goal understanding, leading to better success rates and generalization in complex tasks.

Contribution

The paper proposes GACE loss and GDAN, novel methods that enable agents to better discriminate and focus on goals in multi-target environments using self-supervised learning.

Findings

GDAN outperforms state-of-the-art in success ratio

Improves sample efficiency and generalization

Enhances goal-directed focus in agents

Abstract

Learning in a multi-target environment without prior knowledge about the targets requires a large amount of samples and makes generalization difficult. To solve this problem, it is important to be able to discriminate targets through semantic understanding. In this paper, we propose goal-aware cross-entropy (GACE) loss, that can be utilized in a self-supervised way using auto-labeled goal states alongside reinforcement learning. Based on the loss, we then devise goal-discriminative attention networks (GDAN) which utilize the goal-relevant information to focus on the given instruction. We evaluate the proposed methods on visual navigation and robot arm manipulation tasks with multi-target environments and show that GDAN outperforms the state-of-the-art methods in terms of task success ratio, sample efficiency, and generalization. Additionally, qualitative analyses demonstrate that our proposed method can help the agent become aware of and focus on the given instruction clearly, promoting goal-directed behavior.