This past summer, I conducted research in the Assistive Agent Behavior and Learning Lab, where I studied the sim-to-real gap in robotics’ simulations. I investigated how digitally replicated environments often fail to accurately mirror the behavior of real robotic systems and explored how human feedback can refine a robot’s simulations, ensuring its virtual performance more closely aligns with reality. I experimented with depth cameras, OpenCV, and 3D reconstruction libraries to scan real objects and import them into the Genesis simulator, documenting the challenges of mesh accuracy, missing depth detail, and the limitations of models that only handle rigid objects. Using Iterative Residual Tuning and TuneNet, I learned that previous researchers have approached this problem and identified gaps, particularly in deformable materials like clay and in human-centered evaluation.