Artificial Intelligence Hub

Hey guys, DeepMAV.ai team here.

We are excited to announce the first release of our AI Hub service with the deployments of version 0.3 of Drone Cloud and Drone Swarm products to production.

AI Hub in Drone Cloud

The AI Hub service consists of two main components: Machine Vision and Data Explorer.

Machine Vision allows the creation and running of deep neural net machine vision models in the cloud. New models can be configured based on the built-in YOLOv5 model architecture. Drone Cloud also allows the configuration of custom-trained models based on the same architecture. Vision models can then be used to perform inference on images stored in data sets in Drone Cloud. You can use images captured by the drones on survey missions or any other manually uploaded images. Vision tasks are performed as background scheduled jobs in Drone Cloud on our GPU cluster.

The Data Explorer is our new solution for working with tabular data in Drone Cloud. It allows you to define new tabular data sets, edit tables and data rows, as well as view data captured by the drones in existing data sets. Data Explorer is used in conjunction with Machine Vision to define inputs for vision tasks and review vision inference outputs.

Inference image available in Drone Cloud AI Hub service

AI Hub in Drone Swarm

AI Hub service brings many new drone capabilities as well. The key new capability included in AI Hub is drone-based machine vision. The Drone Swarm equipped drones are now capable of performing real-time machine vision using the camera sensor and can utilize it for target discovery, target position estimation, and dynamic navigation in survey missions. Same machine vision models used in Drone Cloud can automatically be deployed to Drone Swarm for onboard machine vision.

Dynamic survey navigation supports several different modes. In dynamic orbiting, a drone will use vision to discover the target, estimate its position, and then perform orbiting of the discovered target. In the traverse dynamic mode, the drone will traverse a given flight plan looking for all given targets along the way. In the random walk dynamic pattern, the drone will perform a series of random translations and heading changes looking for a given target. We hope to add many more dynamic navigation mission patterns in the future.

Drone Swarm now also supports real-time online machine vision via the Drone Cloud vision REST API. While on mission, drones connected to Drone Cloud can leverage the power of GPU cloud to perform extra large model-based inference at lightning speed. Drones will utilize the inference output from online, cloud-based machine vision same as from the onboard vision – for dynamic navigation.

Imagining New Applications

Our product suite just got a major enhancement with the addition of the AI Hub service. We were already able to tackle important applications in energy & resources such as facility survey inspections. Now, with the addition of dynamic target discovery and position estimation used in survey missions, we are capable of much more. Our products can be utilized for search & rescue, security, smart farming, and other applications where machine vision is critical. We now also have the ability to do autonomous feature discovery in surveyed images, removing the need for tedious manual work.

We will continue to enhance our products with new features and imagine new applications in the future. Our priorities remain energy & resources, agriculture, security, and environmental sectors. We are excited to see our products tested in the field in the next year.