The gap between simulation and reality (sim2real) remains one of the central challenges in deploying learning-based robotic systems in the real world. While simulation provides a scalable, safe, and cost-efficient environment for training and testing, models often fail to generalize when transferred to real-world scenarios due to differences in dynamics, perception, and environmental variability. Recent advances in domain randomization, adaptation, and sim-enhanced data generation have demonstrated promising progress, yet robust and reliable sim2real transfer is far from solved. This workshop aims to convene researchers working at the frontier of simulation and real-world robotics to share advances, challenges, and open problems. Example topics of interest include high-fidelity simulation of deformable objects, realistic contact dynamics, advances in tactile and force sensor simulation, and generalizable sim2real approaches in manipulation. By bringing together experts from these diverse areas, we aim to foster cross-pollination of ideas, establish benchmarks and evaluation criteria, and identify pathways toward unifying principles for robust sim2real transfer. Ultimately, the workshop will serve as a platform for advancing both theoretical understanding and practical deployment, bridging the divide between simulation and reality.
Topics of Interest include but not limited to:
High-fidelity simulation of deformable objects, fluids, and other complex materials
Advances in tactile sensing simulation and multimodal sensor modeling
Sim2real transfer for manipulation, locomotion, loco-manipulation
Domain randomization and domain adaptation techniques for robust transfer
Leveraging synthetic data generation for perception and control
Benchmarks, datasets, and evaluation metrics for measuring sim2real progress
Hybrid pipelines combining simulation and real-world data
Methods for closing the dynamics gap (e.g., contact modeling, physics fidelity)
Real-time and scalable simulators for training large-scale models
Synthetic data generationÂ
Failure cases and negative results in sim2real transfer