Joshua Tobin

Papers

Total Citations

208

H-Index

About

Joshua Tobin’s research lies at the intersection of robotics, computer vision, and deep learning, with a central focus on bridging the “reality gap” between simulation and the real world. His most influential work, “Transfer from Simulation to Real World through Learning Deep Inverse Dynamics Model” (2016, 167 citations), pioneered methods for training control policies in simulation and transferring them to physical robots—a critical step toward safer, more scalable robotic learning. He further advanced this area by exploring domain randomization and generative models for robotic grasping (2018, 24 citations), demonstrating how synthetic data can improve generalization across diverse objects. In “Geometry-Aware Neural Rendering” (2019, 12 citations), Tobin extended neural rendering techniques to capture complex 3D structures, enhancing robotic perception. His doctoral thesis (2019, 5 citations) synthesized these contributions, offering a comprehensive framework for using synthetic data to train real-world robotic systems. Tobin’s work has been instrumental in making simulated training a viable path for real-world deployment, reducing the need for costly physical data collection. His research continues to inspire new approaches in sim-to-real transfer, domain adaptation, and 3D scene understanding.