Reward-Based Environment States for Robot Manipulation Policy Learning

Cédérick, Mouliets, Ferrané, Isabelle and Cuayahuitl, Heriberto (2021) Reward-Based Environment States for Robot Manipulation Policy Learning. In: NeurIPS 2021 Workshop on Deployable Decision Making in Embodied Systems (DDM).

Documents
Reward-Based Environment States for Robot Manipulation Policy Learning
Authors' Accepted Manuscript
[img]
[Download]
[img]
Preview
PDF
cm_NeurIPS2021.pdf - Whole Document

298kB
Item Type:Conference or Workshop contribution (Poster)
Item Status:Live Archive

Abstract

Training robot manipulation policies is a challenging and open problem in robotics and artificial intelligence. In this paper we propose a novel and compact state representation based on the rewards predicted from an image-based task success
classifier. Our experiments—using the Pepper robot in simulation with two deep reinforcement learning algorithms on a grab-and-lift task—reveal that our proposed state representation can achieve up to 97% task success using our best policies.

Keywords:robot manipulation, deep reinforcement learning
Subjects:G Mathematical and Computer Sciences > G700 Artificial Intelligence
G Mathematical and Computer Sciences > G760 Machine Learning
H Engineering > H671 Robotics
Divisions:College of Science > School of Computer Science
ID Code:47522
Deposited On:20 Jan 2022 15:10

Repository Staff Only: item control page