Augmented Reality platform for industrial process automation

Overview

Client

Internal R&D project partially funded by The National Centre for Research and Development

Scope of work

Developing deep learning algorithms for 6D object pose estimation, speech enhancement, keyword spotting and gesture recognition to be used on our AR platform for industrial processes automation.

Methods

Deep Neural Networks, Image Processing, Object Detection, Image Segmentation, Object 6D Pose Estimation, Domain Adaptation, Domain Randomization, Perspective-n-Point, Keyword Spotting, Speech Denoising, Hand Detection and Gesture Recognition

Technologies

Python, PyTorch, C++, TensorFlow, OpenCV, Robotic Operating System, Yaskawa and Fanuc robotic arms, MoveIt library, Unreal Engine

Streamlining production processes with AR and ML

Within this internal R&D project, partially funded by The National Centre for Research and Development, we are developing the system for streamlining production processes using augmented reality (AR) and machine learning.

Our solution aims to increase the efficiency of employees working with advanced production machines, facilitate the transfer of knowledge and reduce the level of training required for effective service work. It can also be used as a training platform for new employees.

Read more about our AR platform in nsFlow case study.

Are you ready to take your business to the next level?

Discover how AI solutions can benefit your company. Schedule a free consultation with us today!

NeuroSYS is trusted by

In a nutshell

displaying info about the device within the operator’s field of view
providing instructions regarding cyclical inspections, setups, maintenance, service, etc.
„virtual inspection” of devices with the help of visualised readings from IoT sensors
remote consultation with an expert with the possibility to annotate an augmented image
employee navigation within the work environment

DL algorithms for 6D pose estimation

We develop deep learning algorithms for 6D pose estimation, which can be trained on synthetic data generated based on CAD models, so our solution can be a good fit for clients’ needs. To ensure accuracy of our methods in real-life we collect video recordings using robotic arms (so it is feasible to annotate objects with their 3D position and orientation on every frame).

We also create our own database for speech enhancement and keyword spotting (in English and Polish) to develop our solutions for speech denoising and commands recognition.

See another cool project that we’ve made