Timezone: »

Goal-Aware Cross-Entropy for Multi-Target Reinforcement Learning
Kibeom Kim · Min Whoo Lee · Yoonsung Kim · JeHwan Ryu · Minsu Lee · Byoung-Tak Zhang

Tue Dec 07 04:30 PM -- 06:00 PM (PST) @ Virtual

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.

Author Information

Kibeom Kim (Seoul National University)
Min Whoo Lee (Seoul National University)
Yoonsung Kim (Seoul National University Biointelligence Lab)
JeHwan Ryu (Seoul National University)
Minsu Lee (Seoul National University)
Byoung-Tak Zhang (Seoul National University & Surromind Robotics)

More from the Same Authors