Brian Okorn

I am a senior research scientist, technical lead, and manager at The Robotics and AI Institute, with over 15 years of experience pioneering advancements in robotics perception, 3D vision, and robot learning. My expertise lies at the intersection of deep learning and physical embodiment, specifically designing and deploying large-scale models for robotic manipulation, open-world object detection, and dynamic 3D scene understanding. I have a proven track record of scaling research initiatives from concept to top-tier peer-reviewed publications and robust, real-world robotic deployment.

I earned my PhD from the CMU Robotics Institute (R-PaD Lab), where my research focused on zero-shot and self-supervised learning for pose estimation and scene flow. My career in autonomy includes developing perception algorithms for maritime and subterranean robots at the Naval Information Warfare Center (NIWC Pacific) and engineering vision systems for NASA’s Robonaut under the prestigious NSTRF Fellowship.

Email  /  GitHub  /  Google Scholar  /  LinkedIn  /  Resume

Research

My research centers on object-centric vision for manipulation, with a focus on zero-shot learning, self-supervised methods, and capturing structured information from human demonstrations.

XSemanticFlow: Cross Object Semantic Alignment for Generalizable Manipulation

Show, Don’t Tell: Detecting Novel Objects by Watching Human Videos

Tactile Embeddings for Multi-Task Learning

Learning Interaction Constraints for Robot Manipulation via Set Correspondences

TAX-Pose: Task-Specific Cross-Pose Estimation for Robot Manipulation

Deep Projective Rotation Estimation through Relative Supervision

OSSID: Online Self-Supervised Instance Detection by (and for) Pose Estimation

IFOR: Iterative Flow Minimization for Robotic Object Rearrangement

Object Pose Estimation without Direct Supervision

Self-Supervised Point Cloud Completion via Inpainting

ZePHyR: Zero-shot Pose Hypothesis Rating

Visual Self-Supervised Reinforcement Learning with Object Reasoning

Learning Orientation Distributions for Object Pose Estimation

Cloth Region Segmentation for Robust Grasp Selection

Just Go with the Flow: Self-Supervised Scene Flow Estimation

Learning Adaptive Sampling Distributions for Motion Planning by Self-Imitation

Counter Tunnel Exploration, Mapping, and Localization with an Unmanned Ground Vehicle

Ego-Motion Estimation on Range Images using High-Order Polynomial Expansion

Smuggling Tunnel Mapping using Slide Image Registration

Toward Automated Modeling of Floor Plans

Using Laser Scanners for Modeling and Analysis in Architecture, Engineering, and Construction

Design and source code from Leo Keselman's website, originally from Jon Barron's website