I'm Tim J. Schoonbeek

We propose to leverage temporal features towards procedure step recognition, significantly reducing the delay between correctly completed procedure steps and their subsequent recognition.

We propose a supervised representation learning approach for generalizable assembly state recognition in manufacturing environments. Our method learns robust features that can generalize across different assembly scenarios and conditions.

We present a novel approach for automatically detecting and segmenting assembly errors in industrial settings. Our method can identify various types of mistakes during manufacturing processes, improving quality control.

We introduce IndustReal, a comprehensive dataset for procedure step recognition in egocentric videos. We also present the new Procedure Step Recognition (PSR) task and evaluation metrics.

We explore methods to extract meaningful procedural information beyond simple action recognition from video recordings. Our approach provides richer understanding of complex manufacturing and maintenance procedures.

We present a proof-of-concept augmented reality system that automatically generates standardized logs for manufacturing and maintenance procedures. The system reduces manual logging overhead while improving consistency and searchability.

We develop a machine learning approach to predict collision risk from simulated video data. Our method demonstrates effective sim-to-real transfer for autonomous driving safety applications.