Multi-AUV Testbed for SLAM and Navigation Research at UMich

Vehicle Overview

PeRL has modified two commercial Ocean-Server Iver2 AUV systems for simultaneous localization and mapping (SLAM) research at the University of Michigan (UMich). PeRL upgraded the vehicles with additional navigation and perceptual sensors including 12-bit stereo down-looking Prosilica cameras, a Teledyne 600 kHz RDI ExplorerDVL for 3-axis bottom-lock velocity measurements, a KVH single-axis fiber-optic gyroscope for yaw rate, and a WHOI Micromodem. To accommodate the additional sensor payload, a new Delrin nosecone was designed and fabricated in-house. Additional 32-bit embedded CPU hardware was added for data-logging, real-time control, and in-situ real-time SLAM algorithm testing and validation. The vehicles, as shipped, are rated to a maximum depth of 100m and a maximum speed of approximately 4 knots. The standard vehicle weighs 29.48kg and can be transported by two people.

PeRL's multi-AUV testbed is based upon a modified Ocean-Server Iver2 AUV platform. Shown in the foreground is a modified vehicle displaying a new nose cone designed and fabricated by PeRL. For comparison, shown in the background is a stock vehicle displaying the original nose cone.

Research Overview

Engineering Science Research The engineering research focus is multi-vehicle multi-scalar simultaneous localization and mapping (SLAM) – the capability for a robot(s) to map an unknown environment while simultaneously using that map to navigate. The algorithmic advancements developed and tested using the UMich Iver2 AUV platforms will advance the multi-resolution navigation and mapping capabilities of robotic AUVs in benthic environments. The experimental navigation framework will fuse navigation information from three disparate technologies, namely: real-time vision-based seafloor navigation, for high-resolution micro-scale navigation, with inertial and acoustic modem-based navigation, for large-area macro-scale navigation. In pursuit of this goal, the engineering science research focuses on three closely related objectives:

1. real-time visual perception: new methods for feature extraction and robust correspondence establishment in benthic environments;
2. multi-scalar representation: techniques for management of data association uncertainty and navigation error over local and large spatial scales;
3. low-bandwidth distributed estimation: techniques for multi-vehicle cooperative navigation using inter-vehicle ranging and state information communicated through a low-bandwidth acoustic communications link

The combined result is a decentralized navigation methodology that provides inter-nodal ranging and data sharing among heterogeneous nodes over a spectrum of spatial scales. The goal is to improve the localization and mapping ability of disparate platforms by leveraging the perception and localization capability of neighboring vehicles via a (low-bandwidth) distributed estimation framework – thus, improving the precision and scale of robotic mapping in ocean science.

Exploded view of PeRL's custom nose cone housing DVL, stereo cameras, WHOI modem, Desert-Star pressure sensor, and Microstrain AHRS.

Field Experimentation The Iver2 platform was chosen (and modified) to serve as a convenient, cost-effective platform for research, development, and experimental validation of vehicle control systems and navigation techniques. PeRL has teamed up with the NOAA Thunder Bay National Marine Sanctuary to use these experimental vehicles to perform exploratory navigation and mapping in the Great Lakes. Two levels of mapping granularity will be provided with these vehicle platforms: (1) large-area survey coverage via sidescan sonar from mid-water column surveys, and (2) high-resolution optical imaging surveys from near-bottom surveys. These complementary sensing scales will allow for large-area searches of new wrecks in and around the proposed expansion of Sanctuary borders, while the high-resolution optical surveys will provide visual confirmation of acoustic targets.

Modified Iver2 sensor layout.