ADAPT Multi-Mission Payload

An open source platform for deploying state of the art deep-neural-network computer vision in real time on small unmanned aircraft systems (sUAS).

Get the source code: https://gitlab.kitware.com/adapt/adapt_ros_ws

• Optimized drone-based collection of imagery and geospatial metadata with live feedback to maintain quality control.

• Integration with the open source do-it-yourself AI toolkit VIAME to annotate data and train mission-specific image-processing models.

• Upload your models for aerial deployment with real-time, georegistered analytics wirelessly transmitted to a ground station computer and beyond for rapid dissemination.

• Commodity hardware components, CAD models, and open-source software allows organizations to cheaply and easily build their own payloads

Real-Time Ground-Station View of Remote Payload Operation

Real-Time Deep Neural Network Ice Segmentation

Frozen Water (Blue) vs. Background (Green)

Supports a variety of unique missions

• Sea and River Ice Monitoring
• Monitoring Arctic Mammal Populations
• Person Search and Rescue
• Wild Fire Monitoring
• Coastline Erosion Monitoring

Ongoing work on the ADAPT project is funded by NOAA to support key missions.

Explore

The ADAPT payload source code is hosted here: https://gitlab.kitware.com/adapt/adapt_ros_ws or try the simulator with docker

Events

• Kitware and ACUASI September 2021 data collection in Fairbanks Alaska.
• The 3rd NOAA Workshop on Leveraging AI in Environmental Sciences

Papers / Presentations

• (2022) National Innovation Center Seminar: Slides

• (2022) Ocean Sciences Meeting: Slides

• (2022) 1st International Workshop on Practical Deep Learning in the Wild at AAAI Conference on Artificial Intelligence: Paper

• (2021) NOAA Innovators Series: Slides, Recording

• (2021) The 3rd NOAA Workshop on Leveraging AI in Environmental Sciences: Slides, Recording