RL3D: Visual Reinforcement Learning with Self-Supervised 3D Representations

Abstract

A framework for visual reinforcement learning using pretrained 3D visual representations for robust sim-to-real transfer.