Things Sam does.
-
Google Research and Google X Moonshots
R&D Software Engineer 2016 - 2020
Created Google’s ‘Frankenphone’, a novel ground-truth multi-camera capture system used to collect 50,000 images that trained the machine learning model used in the product “Learned Portrait Mode.”
This product was launched on Google Pixel phones in 2018, directly making the Google phone brand the leader in mobile photographic image quality.
-
Amazon Robotics AI
Sr. Robotics Software Engineer 2020 - 2021
Team Owner (~10 person) - Prototyping novel robotic systems for optimizing Amazon warehouses via mechatronics / robotic arms. Applied CV/ML. The 2020 robotic system achieved a 200% pick speed improvement, top of the competition.
-
Wireless Software Synchronization of Multiple Distributed Cameras
In Google Research, led the creation of open-source `libsoftwaresync`: A library for wirelessly synchronized multi-smartphone photography, used in the Google product “Portrait Mode”.
Published & patented related tech for multi-camera synchronization and ML.
-
SpaceX
Avionics Flight Software Intern 2013
Developed five critical flight GNC algorithms for the Falcon 9 v1.1 rocket, as well as a larger, more robust telemetry UI for ground control.
Launched September, 2013.
-
Patents for developing a novel solid-state structured light projector system for 3D pose estimation.
US10212785B2 - Staggered array of individually addressable light-emitting elements for sweeping out an angular range.
US10139217B1 - Array based patterned illumination projector.
US9909862B2 - Curved array of light-emitting elements for sweeping out an angular range.
-
FLIR Android Thermal Cameras
Liaison between Google X Moonshots - FLIR’s development of the first FLIR Android thermal camera prototypes on two android devices.
Met with VPs demonstrating a new product concept and handled mobile integration of prototypes as well as the tech transfer handoff.This product launched in FLIR One Gen 1 and onwards.
Personal Projects & Education
-
-
Automated Warehouse Simulation
Created a full-stack simulation of a 2D grid-based automated warehouse with inventory management system and a system of robots completing orders. Deployed to Kubernetes cluster.
Multi-Agent PathFinding (MAPF) for 2D Robots moving inventory on a grid - Building a simulation environment + robots + planning + inventory management system etc.
-
Tensorflow Chessbot
Designed code and educational tutorials for chessboard detection and chess piece classification using DNNs + classic CV.
-
PruDAQ
Co-Created a Google Open-Sourced DAQ board.
PRUDAQ is an open source 40MSPS (megasamples per second) Data Acquisition board for the Beaglebone Black/Green
-
Tensorflow Chessboard Detection
Created an experimental real-time video application that identifies and tracks chessboards in video streams at 30-200ms per frame. It uses TensorFlow for detection, OpenCV for video capture, and Lucas-Kanade tracking, optimized through RANSAC. Initially developed in Python and C++/Halide, the project focuses on Python due to build issues, offering opportunities for further performance enhancements.