Sim-to-Real Robot Learning from Pixels with Progressive Nets

TL;DR

This paper introduces a progressive network framework that effectively transfers deep reinforcement learning policies from simulation to real robots using raw visual input, overcoming the reality gap in pixel-based robot control tasks.

Contribution

The authors propose a novel progressive network approach that enables transfer of learned policies from simulation to real robots, facilitating end-to-end visual control without model-based optimization.

Findings

Successful real-world robot manipulation from raw pixels.

Effective transfer of policies from simulation to real robot.

Deep reinforcement learning with sparse rewards suffices for task learning.

Abstract

Applying end-to-end learning to solve complex, interactive, pixel-driven control tasks on a robot is an unsolved problem. Deep Reinforcement Learning algorithms are too slow to achieve performance on a real robot, but their potential has been demonstrated in simulated environments. We propose using progressive networks to bridge the reality gap and transfer learned policies from simulation to the real world. The progressive net approach is a general framework that enables reuse of everything from low-level visual features to high-level policies for transfer to new tasks, enabling a compositional, yet simple, approach to building complex skills. We present an early demonstration of this approach with a number of experiments in the domain of robot manipulation that focus on bridging the reality gap. Unlike other proposed approaches, our real-world experiments demonstrate successful task learning from raw visual input on a fully actuated robot manipulator. Moreover, rather than relying on model-based trajectory optimisation, the task learning is accomplished using only deep reinforcement learning and sparse rewards.