Modular Deep Q Networks for Sim-to-real Transfer of Visuo-motor Policies

TL;DR

This paper presents a modular deep reinforcement learning approach that enables effective transfer of visuo-motor policies from simulation to real robots, reducing the need for extensive real-world data.

Contribution

It introduces a modular architecture with a perception-control bottleneck, allowing independent training and fine-tuning for improved sim-to-real transfer in robotic tasks.

Findings

Achieved 1.6 pixel accuracy after fine-tuning on a real robot

Significantly outperformed naive transfer with 17.5 pixels error

Demonstrated potential for broader applications in robotic visuo-motor learning

Abstract

While deep learning has had significant successes in computer vision thanks to the abundance of visual data, collecting sufficiently large real-world datasets for robot learning can be costly. To increase the practicality of these techniques on real robots, we propose a modular deep reinforcement learning method capable of transferring models trained in simulation to a real-world robotic task. We introduce a bottleneck between perception and control, enabling the networks to be trained independently, but then merged and fine-tuned in an end-to-end manner to further improve hand-eye coordination. On a canonical, planar visually-guided robot reaching task a fine-tuned accuracy of 1.6 pixels is achieved, a significant improvement over naive transfer (17.5 pixels), showing the potential for more complicated and broader applications. Our method provides a technique for more efficient learning and transfer of visuo-motor policies for real robotic systems without relying entirely on large real-world robot datasets.