OmniABiD: Evaluating Sim2Real Transferability in Safety and Risk Monitoring of Human-Robot Collaboration using NVIDIA Omniverse

Modern factories and production lines undergo extensive virtual simulation before their actual construction and operation. Traditionally, the Sim2Real gap has posed significant challenges, particularly in Risk Assessment (RA) and Risk Monitoring (RM). However, recent advancements have substantially reduced this gap. We argue that it has now narrowed sufficiently to enable effective testing of RM methods and validation of safety sensor setups within simulation environments before real-world deployment. To demonstrate this, we use a systematic pipeline for collecting RM data in Human-Robot Interaction (HRI) scenarios, seamlessly applicable in both simulated environments—using NVIDIA Omniverse Isaac Sim—and real-world settings. Furthermore, we introduce the OmniABiD dataset, which encompasses a diverse range of simulated and real-world HRI scenarios, including both safe interactions and hazardous scenarios. This dataset serves as a foundation for validating RM methods. Finally, we compare RM results obtained from both simulated and real-world data to support our hypothesis. Our analysis highlights remaining Sim2Real discrepancies while demonstrating the feasibility and benefits of integrating RM analysis early in the planning and design phases of industrial systems.

Please note that the dataset is provided for research purposes only and should not be used for commercial purposes.