Sant Cugat del Vallès, Catalonia, Spain
Implemented perception algorithms for Stretch, a case-moving robot for logistics applications. 1. Developed a 15-minute automatic fieldable calibration routine for the on-robot camera suite and arm encoder offsets that results in a closed-loop error of < 1cm in the working section of the arm. 2. Implemented various classical perception algorithms for extracting features and understanding the environment using lidars, time-of-flight depth sensors, and RGB cameras. 3. Supported the machine learning team with its training and inference infrastructure and integrated the outputs of the ML models with the larger perception environment. 4. Developed initial ML models for object detection and scene understanding.
Develop and deploy machine learning models for automated Non-Destructive Testing (NDT) inspection and reporting systems. 1. Developed models for automated defect detection in X-ray and other NDT imaging modalities, including complete data acquisition and cleaning pipelines for training robust inspection systems. 2. Supported product development with backend engineering contributions and providing technical mentorship to team members in software engineering best practices.
Formerly the Boston Dynamics AI Institute Researched and developed approaches to learning complex tasks and behavior for robots from videos of human demonstration of tasks. 1. Developed a large multi-modal transformer-based model for task understanding. Based on videos of people demonstrating simple repair of mechanical objects such as bikes, engines, etc, identify salient objects, image regions, and relationships between salient objects required for task completion. 2. Developed a large weakly-supervised DETR-based sequence-to-sequence model for object detection. 3. Developed the software and machine learning infrastructure needed to work within a large interdisciplinary company, including efficient and reliable inference-time on-robot access to learned models. As Software Lead for the Watch-Understand-Do (learning from human demonstration) project team, responsible for developing and maintaining a high level of software engineering quality across the team and Institute. 1. As one of the early employees of the Institute, set standards and procedures for interactions between teams within the Institute. 2. Mentorship of newer employees and interns. This included supervising interns as well as formal mentorship roles for early-career research and applied scientists.
Researched and developed industrial computer vision tools for manufacturing and logistics applications. 1. Designed and implemented PatMax3D, a point cloud alignment algorithm for 3D object localization in industrial environments. 2. Developed classical machine learning models including SVMs and decision trees for high-speed image classification in production lines. 3. Created robust vision algorithms for defect detection, OCR, and dimensional measurement in manufacturing quality control systems. 4. Relocated to Girona office (2018-2019) to drive technical integration of acquired computer vision startup into Cognex product ecosystem.