Benchmarking Domain Randomisation for Visual Sim-to-Real Transfer

11/13/2020
by   Raghad Alghonaim, et al.
9

Domain randomisation is a very popular method for visual sim-to-real transfer in robotics, due to its simplicity and ability to achieve transfer without any real-world images at all. But a number of design choices must be made to achieve optimal transfer. In this paper, we perform a large-scale benchmarking study on these choices, with two key experiments evaluated on a real-world object pose estimation task, which is also a proxy for end-to-end visual control. First, we study the quality of the rendering pipeline, and find that a small number of high-quality images is superior to a large number of low-quality images. Second, we study the type of randomisation, and find that both distractors and textures are important for generalisation to novel environments.

READ FULL TEXT

page 4

page 5

page 6

research
03/23/2022

PRTT: Precomputed Radiance Transfer Textures

Precomputed Radiance Transfer (PRT) can achieve high quality renders of ...
research
04/05/2022

ObjectFolder 2.0: A Multisensory Object Dataset for Sim2Real Transfer

Objects play a crucial role in our everyday activities. Though multisens...
research
12/03/2018

Towards Accurate Task Accomplishment with Low-Cost Robotic Arms

Training a robotic arm to accomplish real-world tasks has been attractin...
research
06/15/2023

MLonMCU: TinyML Benchmarking with Fast Retargeting

While there exist many ways to deploy machine learning models on microco...
research
03/03/2022

Sim2Real Instance-Level Style Transfer for 6D Pose Estimation

In recent years, synthetic data has been widely used in the training of ...
research
06/20/2018

Synthesizing Diverse, High-Quality Audio Textures

Texture synthesis techniques based on matching the Gram matrix of featur...
research
09/28/2021

Image scaling by de la Vallée-Poussin filtered interpolation

We present a new image scaling method both for downscaling and upscaling...

Please sign up or login with your details

Forgot password? Click here to reset