Adapting control policies from simulation to reality using a pairwise\n loss

2018/07/26 by Ulrich Viereck, Viereck, Ulrich, Xingchao Peng +5
Computer Science · #Artificial Intelligence (cs.AI) #Computer Vision and Pattern Recognition (cs.CV) #Domain Adaptation and Few-Shot Learning #FOS: Computer and information sciences #Multimodal Machine Learning Applications #Reinforcement Learning in Robotics #Robotics (cs.RO)

paper · pdf · doi:10.48550/arxiv.1807.10413

openalex publication_date 2018/07/26 · openalex created_date 2022/08/04 · openalex updated_date 2026/07/28

Abstract

This paper proposes an approach to domain transfer based on a pairwise loss\nfunction that helps transfer control policies learned in simulation onto a real\nrobot. We explore the idea in the context of a 'category level' manipulation\ntask where a control policy is learned that enables a robot to perform a mating\ntask involving novel objects. We explore the case where depth images are used\nas the main form of sensor input. Our experimental results demonstrate that\nproposed method consistently outperforms baseline methods that train only in\nsimulation or that combine real and simulated data in a naive way.\n

Related

• Generalizable Domain Adaptation for Sim-and-Real Policy Co-Training
• Sim-to-Real Transfer of Accurate Grasping with Eye-In-Hand Observations and Continuous Control
• Self-Supervised Sim-to-Real Adaptation for Visual Robotic Manipulation
• VR-Goggles for Robots: Real-to-sim Domain Adaptation for Visual Control
• Know Thyself: Transferable Visual Control Policies Through Robot-Awareness