Control Algorithms for UUV Teams Using Acoustic Communications

Introduction: Interest in the use of multiple unmanned underwater vehicles (UUVs) for military and commercial uses is growing because of the potential benefits to underwater operations such as searching, inspection, and surveying. NRL researchers are developing control, communications, and positioning methods to enable the use of multiple UUV formations. In the undersea environment, traditional methods of communication and navigation (radio and GPS) are ineffective. Properties of seawater make control of UUVs very difficult. Because inertial-based vessel positioning systems typically yield position error growth on the order of 1% of the distance traveled,1 they are not adequate for formation maneuvering. The only viable alternative underwater for communications and positioning is acoustics, but these systems yield fairly short ranges and very low bandwidths.2 A promising approach, which is the focus of this work, is vessel relative positioning and navigation using combined communication/position acoustic systems. Formation maneuvering based on inter-vessel positioning and navigation has distinct advantages in that it can reduce or eliminate the requirement for pre-deployed positioning systems, and it can be used to increase the sensor footprint in searching and surveying tasks.

To control multiple vehicle formations, classic logic, behavior-based, and neural network controllers have been developed. The formations are fashioned after those observed in nature, most notably lines of ducks and caterpillars. The classic logic control systems used were modeled after Braitenburg machines and served as a baseline capability for comparison. Neural network controllers, which hold the promise of real-time adaptability, were grown using a genetic algorithm and in situ sensor data. Both computer simulations and tests using mobile land robots equipped with frequency multiplexed acoustic communication systems have demonstrated the feasibility of these approaches. Several real-world problems have been tackled, including acoustic sensor directivity and reverberation.

The algorithms in this work are based on the leader/follower paradigm, commonly used in military operations, in which all of the robots in the formation position themselves relative to an assigned lead robot. The overall lead robot is typically controlled by an operator or programmed to do waypoint following. Figure 1 is a conceptual view of a line formation. In this figure, the green robot, whose identification (ID) is 0, is the lead robot. It is followed by the robot whose ID is 1, which is in turn followed by the robot whose ID is 2, and so on. In this illustration, neural networks control the follower robots while an operator maneuvers the lead robot.

Figure 13: The system operates in a passive manner, meaning that the robots do not exchange position or bearing and range information. Instead, each robot, except the leader, steers itself toward an acoustic chirp emitted from the robot in front of it. For example, the leader chirps at frequency range A, the robot directly behind it steers in the direction that it perceives is the source of the chirp in frequency range A, and at the same time chirps in frequency range B. The robot behind it steers towards the source of frequency range B and so on.

FIGURE 13
Robot line. The green robot is controlled manually. Using their sensors and controllers, the other robots follow.

Figure 14: This figure shows robots in laboratory following each other. As in Fig. 13, the lead robot is controlled manually, and the others follow using sensors and control algorithms. Using the classic logic method, the robot simply steers in the direction of the strongest signal. This method is robust and works well in both simulation and laboratory settings, but it has problems when the source is directly in front or behind. To alleviate this problem, the behavior routine was developed. It alternates between search, seek, and follow modes, depending on whether the signal is getting more or less intense over time. In simulation it works well, but because of reverberation and multipath problems in the laboratory, it is less effective (as illustrated in Fig. 15). The neural network approach relies on a feed-forward neural network trained in a teaching mode to detect the signal direction in relation to the robot. This same network is then used as a controller. Its performance is similar to that of the classic logic technique in both simulation and in laboratory testing.

FIGURE 14
A three-robot following test. The lead robot is being operated manually while the second and third robots are using their microphones to track the robots in front of them.

Figure 15: In concept, this work has shown that formation maneuvering using acoustic sensors is possible in both a simulated environment and in the physical world of the laboratory. It has also shown that the structure of the formation can remain viable without using a centralized controller, or by using external infrastructure-based communications. Since the communications between the robots uses very low bandwidth acoustic methods, this work is relevant to UUV team operations.

FIGURE 15
The wall is reflecting sound back to the robot that would normally be radiated elsewhere. In this case, the robot may read a higher intensity value on its right side, causing it to errantly turn toward the wall.

[Sponsored by ONR]