SBIR-STTR Award

SunBeam: Automatic Postprocessing of Underwater Mapping Data Phase II
Award last edited on: 7/26/2022

Sponsored Program
SBIR
Awarding Agency
DOD : DARPA
Total Award Amount
$1,724,285
Award Phase
2
Solicitation Topic Code
HR001120S0019-04
Principal Investigator
Kristof Richmond

Company Information

Sunfish Inc

3511 Caldwell Lane
Del Valle, TX 78617
   (319) 321-7380
   N/A
   www.sunfishinc.com
Location: Single
Congr. District: 35
County: Travis

Phase I

Contract Number: HR001121C0067
Start Date: 12/14/2020    Completed: 10/21/2021
Phase I year
2021
Phase I Amount
$224,628
We propose the SunBeam underwater data postprocessor to provide rapid, automated generation of large scale environmental models from a wide variety of underwater sensors. This automated software suite will leverage over 15 years of experience at Sunfish, Inc. and Stone Aerospace with advanced, fully 3-dimensional (3D) Simultaneous Localization and Mapping (SLAM) systems used as part of Unmanned Underwater Vehicle (UUV) onboard navigation systems. Currently, processing of seabed data into large-scale 3D models is prohibitively resource intensive. There are many methods to obtain large underwater data sets using a variety of platforms and sensing modalities. This data contains a variety of systematic errors, random noise, and outlier measurements. State-of-the-art processing of such data relies on trained human operators both to clean the data and optimize automated processing parameters. No generalized, automated framework for generation of 3D models from underwater data currently exists. Finally, state-of-the-art underwater environmental models are typically not fully 3D and generally achieve resolutions on the order of 1e-9 of the mapped volume. With SunBeam, we propose to scale our existing processing algorithms—which can already generate full 3D environmental models on small-scale multibeam sonar datasets with modest human intervention—to a system which can incorporate a wide variety of long- and short-range underwater sensors. We will assimilate data on the scale of 100 km2, 1500 m vertical variation, and 5 mm resolution, and fully automate the processing of an environmental model within 2 days. Phase I will investigate high-risk portions of our approach to determine its feasibility, and develop a prototype implementation, which will incorporate multibeam sonar and monocular camera data from a variety of sensor manufacturers. Our approach will leverage our current fully-3D processing system which is already optimized to rapidly (up to real time) produce environmental models used for UUV localization, planning, and simulation. We will also leverage our existing SUNFISH UUV which can autonomously explore and map complex 3D underwater environments using a multibeam sonar, inertial sensors, Doppler velocity log, and high definition camera. This platform will enable us to generate new camera and sonar datasets for testing and training the prototype implementation, allowing for rapid, iterative development of the technology. The two highest-risk elements of this development are (1) scaling the data, processing, and output model to resolutions of 1e-18 of the mapped volume and (2) automating processing to completely remove humans from the loop. To generate a continuous, high-resolution 3D model, both long and short range sensors must be used, and gaps in the model must be filled. Our Phase I development plan will focus on the following four technical objectives: adding new sensing modalities, automation, computational scaling, and filling in gaps.

Phase II

Contract Number: HR001122C0065
Start Date: 2/25/2022    Completed: 2/24/2025
Phase II year
2022
Phase II Amount
$1,499,657
Currently, processing of seabed data into large-scale 3-dimensional (3D) models suitable for simulation and mission planning is prohibitively resource intensive. There are many methods to obtain large underwater data sets using a variety of platforms and sensing modalities but this data contains many systematic errors, random noise, and outlier measurements. Additionally, state-of-the-art processing systems that exist typically do not fuse multiple sensors into a single 3D model suitable for simulation & mission planning. These existing systems also rely on trained human operators both to clean the data and optimize automated processing parameters. Finally, state-of-the-art underwater environmental models are typically not fully 3D and generally achieve resolutions of 150 m3 which is 1e-9 orders of magnitude less than the desired volume. Being able to rapidly generate large 3D maps that can address the above limitations will enable many applications for defense, science, infrastructure, recreation and exploration. Sunfish Inc. proposes the SunBeam underwater data postprocessor to provide rapid, automated generation of large scale environmental models suitable for simulation, free of any gaps and holes, from a wide variety of underwater sensors. We propose to generate full 3D environmental models on small-scale multibeam sonar and camera datasets with modest human intervention—to a system which can incorporate a wide variety of long- and short-range underwater sensors with no human intervention. We will assimilate data on the scale of 100 km2, 1500 m vertical variation, and 5 mm resolution, and fully automate the processing of an environmental model that is suitable for mission planning and simulation within 2 days. This automated software suite will leverage over 15 years of experience at Sunfish, Inc. with advanced, fully automated 3D Simultaneous Localization and Mapping (SLAM) systems with sensor fusion used as part of Unmanned Underwater Vehicle (UUV) onboard navigation systems. We will leverage our existing SUNFISH UUV, which can autonomously explore and map complex 3D underwater environments using a multibeam sonar and high definition camera. This platform will enable us to generate new camera and sonar datasets for testing and training the prototype implementation, allowing for rapid, iterative development of the technology.