2022
DOI: 10.48550/arxiv.2207.10821
|View full text |Cite
Preprint
|
Sign up to set email alerts
|

Rethinking Sim2Real: Lower Fidelity Simulation Leads to Higher Sim2Real Transfer in Navigation

Abstract: If we want to train robots in simulation before deploying them in reality, it seems natural and almost self-evident to presume that reducing the sim2real gap involves creating simulators of increasing fidelity (since reality is what it is). We challenge this assumption and present a contrary hypothesis -sim2real transfer of robots may be improved with lower (not higher) fidelity simulation. We conduct a systematic large-scale evaluation of this hypothesis on the problem of visual navigation -in the real world,… Show more

Help me understand this report

Search citation statements

Order By: Relevance

Paper Sections

Select...
2

Citation Types

0
2
0

Year Published

2023
2023
2023
2023

Publication Types

Select...
1
1

Relationship

0
2

Authors

Journals

citations
Cited by 2 publications
(2 citation statements)
references
References 39 publications
0
2
0
Order By: Relevance
“…We do not know whether sim performance reflects real-world performance, whether design choices that improve sim performance improve real-world performance, or whether sim error modes reflect realworld error modes. Most prior sim-to-real studies in navigation focused on spatial (point goal) navigation and legged locomotion (81)(82)(83)(84)(85)(86), as opposed to semantic navigation. The few other realworld semantic navigation works directly trained on real-world images for outdoor visual goals (87,88) and language instruction following the study of Anderson et al (89).…”
Section: Introductionmentioning
confidence: 99%
“…We do not know whether sim performance reflects real-world performance, whether design choices that improve sim performance improve real-world performance, or whether sim error modes reflect realworld error modes. Most prior sim-to-real studies in navigation focused on spatial (point goal) navigation and legged locomotion (81)(82)(83)(84)(85)(86), as opposed to semantic navigation. The few other realworld semantic navigation works directly trained on real-world images for outdoor visual goals (87,88) and language instruction following the study of Anderson et al (89).…”
Section: Introductionmentioning
confidence: 99%
“…Applying these algorithms to real robots can often result in severe issues or failure not observed in simulation [8], [9]. The issues with the simulation-reality gap and the desire to bridge it is not new [10]- [13]. However, unlike the majority of works that discuss this for simulators of a single robot system, we propose a novel approach that addresses the simulation-reality gap without necessarily attempting to shrink this gap unless deemed necessary.…”
Section: Introductionmentioning
confidence: 99%