The Reality Gap in Robotics: Challenges, Solutions, and Best Practices

TL;DR

The paper formalizes the sim-to-real problem and decomposes the reality gap into $G_{dyn}$ and $G_{perc}$, with a performance gap $G_{perf}$ used to evaluate transfer. It surveys dominant mitigation strategies such as domain randomization, real-to-sim transfer, state and action abstractions, and sim-real co-training across locomotion, navigation, and manipulation. It discusses evaluation metrics and practical simulator design considerations that enable robust transfer, while outlining open problems and future directions. The work aims to guide researchers and practitioners in building scalable, reproducible sim-to-real pipelines and benchmarks.

Abstract

Machine learning has facilitated significant advancements across various robotics domains, including navigation, locomotion, and manipulation. Many such achievements have been driven by the extensive use of simulation as a critical tool for training and testing robotic systems prior to their deployment in real-world environments. However, simulations consist of abstractions and approximations that inevitably introduce discrepancies between simulated and real environments, known as the reality gap. These discrepancies significantly hinder the successful transfer of systems from simulation to the real world. Closing this gap remains one of the most pressing challenges in robotics. Recent advances in sim-to-real transfer have demonstrated promising results across various platforms, including locomotion, navigation, and manipulation. By leveraging techniques such as domain randomization, real-to-sim transfer, state and action abstractions, and sim-real co-training, many works have overcome the reality gap. However, challenges persist, and a deeper understanding of the reality gap's root causes and solutions is necessary. In this survey, we present a comprehensive overview of the sim-to-real landscape, highlighting the causes, solutions, and evaluation metrics for the reality gap and sim-to-real transfer.

Figures (2)

• Figure 1: The reality gap is the difference between a simulated and real environment in aspects related to dynamics, perception & sensing, actuation & control, and system design. By carefully designing these modules, the gap can be reduced to a reasonable size. Sim-to-real as well as real-to-sim transfer require methods that can carefully overcome the remaining reality gap. Figures are adapted from successful sim-to-real applications in various robotics domains akkaya2019solvingtan2018simTang-RSS-23lee2020learningradosavovic2024realkaufmann2023champion.
• Figure 2: Illustration of the different sources of the reality gap in the real world (left) and simulation (right).