AQUA: an aquatic walking robot

Christina Georgiades, Andrew German, Andrew Hogue, Hui Liu, Chris Prahacs, Arlene Ripsman, Robert Sim, Luz Abril Torres-Méndez, Pifu Zhang, Martin Buehler, Gregory Dudek, Michael Jenkin, Evangelos E. Milios

Abstract

The AQUA project introduces a biologically inspired hexapod robot designed for versatile underwater locomotion. By adapting the terrestrial RHex platform with specialized paddles, AQUA achieves both walking on the seabed and swimming in open water. This dual-mode capability addresses the limitations of traditional thruster-based autonomous underwater vehicles (AUVs), particularly in station-keeping and shallow-water operations. This paper describes the robot's mechanical design, its trinocular vision system, and acoustic localization methods used for autonomous navigation.

Introduction

Traditional underwater robots typically rely on thrusters for propulsion. While effective in open water, thruster-based systems struggle with "station-keeping"—maintaining a fixed position—especially near the seabed where thruster wash can disturb silt and obscure sensors. Furthermore, these vehicles often cannot operate in very shallow water or transition seamlessly from land to sea.

The AQUA project addresses the need for a robot capable of Site Acquisition and Scene Re-inspection (SASR). This task requires a robot to autonomously navigate to a specific underwater site (such as a coral reef or a leaking pipe), maintain a stable pose to collect high-quality sensor data over long periods, and return to the site later for follow-up observations.

System Design and Locomotion

The AQUA robot is a derivative of the RHex hexapod platform. Its core innovation lies in its dual-purpose locomotion system, which utilizes six independently controlled paddles.

Figure 2: CAD model of the AQUA robot showing the hexapod structure and paddle design.

In open water, these paddles act as fins or control surfaces, providing five degrees of freedom (surge, heave, pitch, roll, and yaw). When the robot reaches the seabed, the paddles function as legs for walking. Using legs for "walking" provides a level of stability for sensor readings that is nearly impossible for hovering thruster-based robots to achieve without significant power consumption and environmental disturbance.

Sensing and Navigation

To support autonomous operation, the robot carries a custom sensor package and utilizes multiple localization modalities:

  • Trinocular Vision: The robot is equipped with three Firewire CCD cameras and an Inertial Navigation System (INS). This allows for real-time 3D surface reconstruction and the registration of depth maps into a global coordinate system.
  • Acoustic Localization: To navigate when vision is obscured by turbidity or "marine snow," the system uses an array of four hydrophones. By calculating the time-delay of arrival from a sound source, the robot can triangulate its position.
  • Monocular Depth Extrapolation: A Markov Random Field (MRF) model estimates depth from single-camera images by combining intensity data with sparse depth measurements from the stereo head.
Figure 3: The AQUA robot undergoing underwater testing in a controlled pool environment.

Experimental Results

The authors evaluated the system using experimental pool data at McGill University and lab-based INS data. Key results include:

  • Locomotion Success: The robot successfully demonstrated swimming and diving to depths of 10 meters, controlled by an onboard computer and inclinometer.
  • Acoustic Precision: Using a sampling frequency of 44100 Hz, the acoustic system proved capable of estimating the direction of arrival of sound pulses, providing a viable alternative to visual navigation.
  • 3D Modeling: The trinocular system successfully recovered 3D surfaces, though the authors noted that while the INS provides highly accurate orientation, its positional accuracy drifts quickly and requires visual correction.

Conclusion and Future Work

The AQUA project demonstrates that a legged underwater robot can overcome many of the station-keeping and environmental limitations of thruster-based AUVs. Future work will focus on addressing positional drift in the INS system and extending SLAM (Simultaneous Localization and Mapping) capabilities from 2D manifolds to full 3D in unstructured underwater environments. Additionally, further validation in real-world turbid water conditions is required to refine the monocular depth estimation algorithms.

Summary updated

Abstract

This paper describes an underwater walking robotic system being developed under the name AQUA, the goals of the AQUA project, the overall hardware and software design, the basic hardware and sensor packages that have been developed, and some initial experiments. The robot is based on the RHex hexapod robot and uses a suite of sensing technologies, primarily based on computer vision and INS, to allow it to navigate and map clear shallow-water environments. The sensor-based navigation and mapping algorithms are based on the use of both artificial floating visual and acoustic landmarks as well as on naturally occurring underwater landmarks and trinocular stereo.
BibTeX
@inproceedings{georgiades2004aqua,
  author = {Christina Georgiades and Andrew German and Andrew Hogue and Hui Liu and Chris Prahacs and Arlene Ripsman and Robert Sim and Luz Abril Torres-Méndez and Pifu Zhang and Martin Buehler and Gregory Dudek and Michael Jenkin and Evangelos E. Milios},
  title = {{AQUA: an aquatic walking robot}},
  booktitle = {IEEE/RSJ International Conference on Intelligent Robots and Systems},
  year = {2004},
  doi = {10.1109/iros.2004.1389962},
}
Figure