Autonomous Sailboating

Project Overview

We want to get a sailboat (a type of Unmanned Surface Vehicle, or USV) through a canal autonomously. The maneuver will be performed in a simulation environment with simulated wind as the propellant for the USV. Environmental disturbances (water currents) will also be an important factor. This work has useful implications because it allows the deployment of navigating, energy-efficient and green-energy sailing boats that can withstand some environmental disturbances and only use abundant wind energy to propel themselves. These boats could be used in sea monitoring, oceanic measurement, as drifting databuouys, as survey ships etc. This can reduce expenses and reduce deployment time, and increase data resolution.

Implementation and Results

Fig 2: Block diagram of planner and controller

Our control design modularizes the system into two stages: a planner, which takes in high level x,y position targets and generates heading trajectories, and a yaw controller, which generates sail and rudder angles to track the reference heading trajectory output by the planner. Since the planning stage does not need to be run at the same rate as the yaw controller due to the slow speed of the boat, we create an inner feedback loop between the yaw controller and the system dynamics, and an outer feedback loop that incorporates the planner and is run at a lower rate. Our paper can be found here.

Our code is based on that of stda-sailboat-simulator. This simulator uses RK45 to solve an ODE of the sailboat dynamics over time.

Fig. 3, 4: Single obstacle avoidance to reach(100,40) with wind in the direction of sailing and no waves [LEFT] Channel navigation to reach(100, 40) with wind in the direction of sailing and 0.5 m waves [RIGHT]

Using the above path planner and controller, our simulated sailboat successfully traversed a channel to reach the target point (100, 40) with 0.5m waves. Note that the wind is blowing in the direction of sailing.

Conclusion

The simulation was run in several different cases. Waves, obstacles, and different wind vectors were used. The tests were not exhaustive due to the wide variety of environmental conditions possible at sea and also due to the limitations of the simulator. An experiment was deemed successful if the sailboat got close to the target position.

The results demonstrated that the planner was largely successful. Passages as narrow as 4 meters were navigated through. Any narrower gaps caused the planner to return an infeasible path. We used:

Sailing with wind To demonstrate that the path planning and con-trol scheme worked, the target position was set to (100,40,π/4) with wind the in the direction of the intended heading, shown in Fig.3. The wind had amagnitude of5 m/s.
Sailing with crosswind In this experiment, the wind is changed to point at−45degrees, roughly at a right angle to the straight linetrajectory from start to goal. See Figure 4.
Sailing against Wind In this experiment, the wind points at−135 degrees, andthe goal state is(x,y) = (40,40). This poses an interestingproblem since in the real dynamics model there is a nullwhen the boat is heading directly into the wind where the boat has a speed of 0.
Single Obstacle: To demonstrate that the path planning and control scheme worked, the target position was set to(100,40,π/4) with wind the in the direction of the intended heading, shown in Fig.3. The wind had a magnitude of18 023 481 908 m/s. The boat was able to sail around the obstacle.
Channel: Because the obstacle avoidance worked without a dis-turbance, waves were added to the channel traversal experiment
Crosswinds: The path planner was then tested with different wind directionsThe experiment was run with a different wind direction.Note that the sailboat overshoots the target. This was deemed acceptable because stopping a sailboat at a point is another issue

Team Bios

(a) Names and short bios of each member of your group

Neelay is a 4th year undergraduate in EECS, and hastaken courses such as EE 128, 221A/222, and (of course)EECS 106A. His interests are primarily related to controland motion planning

Hamza is a 4th year undergraduate in EECS. He has taken EECS106A,and EECS126. His interests in robotics are in machine learning/ artificial intelligence and dynamically responsive, maneuverable robotic systems with a spice of eﬀiciency/optimization

Andrew is a 4th year undergraduate in EECS. He has taken EE 120, EE C128, and EECS C106A. His interests in robotics are in micro-scale mobile robots.