Brutzman, Don, Healey, Tony, Marco, Dave and McGhee, Bob, "The Phoenix Autonomous Underwater Vehicle," chapter 13, AI-Based Mobile Robots, editors David Kortenkamp, Pete Bonasso and Robin Murphy, MIT/AAAI Press, Cambridge Massachusetts, 1998.
Abstract. The Phoenix autonomous underwater vehicle (AUV) is a robot for student research in shallow-water sensing and control (Figure 1). Phoenix is neutrally buoyant at 387 pounds (176 kg) with a hull length of 7.2 feet (2.2 m). Multiple propellers, thrusters, plane surfaces and sonars make this robot highly controllable. The underwater environment provides numerous difficulties for robot builders: submerged hydrodynamics characteristics are complex and coupled in six spatial degrees of freedom, sonar is problematic, visual ranges are short and power endurance is limited. Numerous Phoenix contributions include artificial intelligence (AI) implementations for multisensor underwater navigation and a working three-layer software architecture for control. Specifically we have implemented the execution, tactical and strategic levels of the Rational Behavior Model (RBM) robot architecture. These three layers correspond to hard-real-time reactive control, soft-real-time sensor-based interaction, and long-term planning respectively. Operational software functionality is patterned after jobs performed by crew members on naval ships. Results from simple missions are now available.
In general, a critical bottleneck exists in AUV design and development. It is tremendously difficult to observe, communicate with and test underwater robots because they operate in a remote and hazardous environment where physical dynamics and sensing modalities are counterintuitive. Simulation-based design using an underwater virtual world has been a crucial advantage permitting rapid development of disparate software and hardware modules. A second architecture for an underwater virtual world is also presented which can comprehensively model all necessary functional characteristics of the real world in real time. This virtual world is designed from the perspective of the robot, enabling realistic AUV evaluation and testing in the laboratory. 3D real-time graphics are our window into the virtual world, enabling multiple observers to visualize complex interactions.
Networking considerations are crucial within and outside the robot. A networked architecture enables multiple robot processes and multiple world components to operate collectively in real time. Networking also permits world-wide observation and collaboration with other scientists interested in either robot or virtual world. Repeated validation of simulation extensions through real-world testing remains essential. Details are provided on process coordination, reactive behaviors, navigation, real-time sonar classification, path replanning around detected obstacles, networking, sonar and hydrodynamics modeling, and distributable computer graphics rendering. Finally in-water experimental results are presented and evaluated.