Climbing Computer Company developed and deployed a computer vision system for a recycling center. When a collected object was not suitable for resale, it was sent on a conveyor to be recycled. The computer vision system continuously tracked how many items were recycled and provided daily reports to management. These reports made it possible to measure and predict equipment uptime, changes in inventory, and income from recycled material.
Client Project
To autonomously collect soil in uncultivated terrain, robotic arms must distinguish between different granular materials and penetrate the correct material. We develop a prototype that collects soil in heterogeneous terrain. If mounted to a mobile robot, it can be used to perform soil collection and analysis without human intervention. Unique among soil sampling robots, we use a general-purpose robotic arm rather than a soil core sampler.
Research Paper
Methods for configuring a quantum annealer to solve a QUBO problem comprises receiving data representing an initial graph representing an embedding of a QUBO problem into a qubit architecture of the quantum annealer and causing one or more GPU thread blocks to create and store a best local current graph and update the best local current graph.
Patent
Learned object detection methods based on fusion of LiDAR and camera data require labeled training samples, but niche applications, such as warehouse robotics or automated infrastructure, require semantic classes not available in large existing datasets. Therefore, to facilitate the rapid creation of multimodal object detection datasets and alleviate the burden of human labeling, we propose a novel automated annotation pipeline. Our method uses an indoor positioning system (IPS) to produce accurate detection labels for both point clouds and images and eliminates manual annotation entirely. In an experiment, the system annotates objects of interest 261.8 times faster than a human baseline and speeds up end-to-end dataset creation by 61.5%.
Research Paper
Robotic access monitoring of multiple target areas has applications including checkpoint enforcement, surveillance and containment of fire and flood hazards. Monitoring access for a single target region has been successfully modeled as a minimum-cut problem. We generalize this model to support multiple target areas using two approaches: iterating on individual targets and examining the collections of targets holistically. Through simulation we measure the performance of each approach on different scenarios.
Research Paper
We present an open-source robotic framework that integrates computer vision and machine learning based inverse kinematics to enable low-cost laboratory automation tasks such as colony picking and liquid handling. The system uses a custom trained U-net model for semantic segmentation of microbial cultures, combined with Mixture Density Network for predicating joint angles of a simple 5-DOF robot arm. We evaluated the framework using a modified robot arm, upgraded with a custom liquid handling end-effector. Experimental results demonstrate the framework's feasibility for precise, repeatable operations, with mean positional error below 1 mm and joint angle prediction errors below 4 degrees and colony detection capabilities with IoU score of 0.537 and Dice coefficient of 0.596.
Research Paper
G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, eg principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
Patent