Robot Perception

Robot perception focuses on enabling robots to sense, interpret, and reason about the physical world through data from cameras, depth sensors, lidar, tactile sensors, and other modalities. Research in this area develops algorithms and learning-based methods that enable robots to build geometric and semantic representations of their environments, including robot pose and proprioception; track objects and agents over time; and understand scene dynamics under uncertainty. Research in this area also develops novel sensors, including tactile sensing and camera-free motion capture.

At Oregon State University, robot perception research spans topics such as visual and multi-modal sensing, 3D reconstruction and mapping, object segmentation, tracking and shape completion, semantic understanding and 3D vision-language models, 3D point tracking, state estimation and world modeling, and perception for manipulation, navigation, and human–robot interaction. Emphasis is placed on robust perception in real-world settings, including unstructured, dynamic, partially observable, and field-robotics environments that may be visibility-limited (e.g., dark operations, turbid water, etc.). This work is tightly integrated with learning, control, and planning to support autonomous robotic systems that can operate reliably in the wild.