Autonomous Stabilization of Boat

The automation technology for a small boat like a fishing boat is not sufficiently investigated yet. After Iwate prefecture was hit by the Great East Japan Earthquake, the number of workers has dropped to 84.3% in Okirai Fisheries Cooperative Association (OFCA). The number of new comers is quite small, because of large body load (1) and difficulty of the task (2). The workers must simultaneously control the boat, rod, and water goggle in order to capture shellfishes like abalones and sea urchins. If the boat will be able to automatically stay a position, problems (1) and (2) will be much relaxed.


In order to realize the autonomous stabilization, we should solve the following problems

  • Measuring the state of boat
  • Controlling the boat considering disturbance (e.g. wave)

Measuring State of Boat

In order to measure the state of a boat, we developed an image processing system. This system calculates the optical flow of the corners of the underwater objects (e.g. stones). These objects are always observable when fishermen capture shellfishes or fishes by their rods. We developed an image processing system that applies optical flow after down sampling, gray scaling, and normalization. This preprocessing is suitable for the underwater images.

Considers Disturbance

Generally, PID control does not provide a good trajectory when a boat is taking external disturbance like wave. We need another control that is based on disturbance. Let us determine the acceleration around a boat as


where a is total acceleration of a boat. a is composed of free dynamics acceleration (water surface resistance, floatation, gravity, and so on) a0,motor acceleration at, and disturbance acceleration aw. We can easily control a boat if we knows all of them.

a is measurable using instrument. In our implementation, we are using underwater image processing and ジャイロ sensor. We can calculate motor acceleration at from the control input of the スラスタ and measured a relatively easily. However, we can not obtain a0 from mathematical equation, because it depends on the shape and velocity of a boat. We applied Dynamics Learning Tree (DLT) for the learning and prediction of a0. When a0 is predictable, we can obtain aw easily.

The following figures show the learning and prediction results of a0. The graphs of real data and predicted data are very close.

Following figures show the used wave generation system and a boat.

Development of Larger Boat

In order to conduct an experiment on the sea, we are developing a larger boat.