Brandon Kinman

Papers

Total Citations

333

H-Index

About

Brandon Kinman is a researcher at the forefront of robotics and computer vision, with a focus on enabling autonomous systems to learn and operate in unstructured, real-world environments. His most impactful contribution is the creation of the **Google Scanned Objects** dataset, a high-quality, open-source collection of photo-realistic 3D household item models. This dataset, which has garnered over 300 citations, directly addresses a critical bottleneck in deep learning: the need for diverse, realistic training data. By providing a large corpus of scanned objects, Kinman’s work has empowered breakthroughs in interactive 3D simulations, allowing robots to train in virtual worlds before acting in the physical one. Building on this foundation, his research into **Demonstration-Bootstrapped Autonomous Practicing** explores how reinforcement learning systems can continuously improve through self-collected data, reducing the need for human intervention. This work paves the way for robots that can autonomously practice and refine skills in messy, real-world settings. Through these contributions, Kinman is helping to bridge the gap between simulated training and robust, real-world robotic performance.