Challenges

A big challenge we encountered throughout this project was the inherent dynamics of our system. The application of a soft underactuated system to a viscous media produces highly nonlinear and nonholonomic phenomenon, which are extremely difficult to predict and model. Accounting for this, we were able to lower our demand for control precision (on the order of several centimeters). Regardless, our ability to apply feedback control to sucessfully stabilize this system in performing motion trajectories is a highly novel and impactful development.

Communicating with the ESP32 while it was inside the fish underwater was also major challenge for us as the WiFi signals were having trouble being received by the ESP32. As a result, we went through a couple iterations of different solutions on tackling the problem, ultimately leading us to keeping the ESP32 floating on top of the water. While this solved our issue, water did leak into the ESP32’s carrier multiple times resulting in it being fried. Another challenge was using a “waterproof” servo motor that can only last for a finite amount of time under water. Throughout the entirety of this project, we went through a minimum of three servo motors as the water eventually causes damage to them.

Future Work

The next stage of our research project is to implement the CV with an occupancy grid to improve the motion planner. Doing so allows us to remove the control points we calibrated ahead of time ensuring that the robotic fish can dynamically find its path to the goal position. For determining the optimal trajectory, we can use algorithms such as A* and RRT to solve for this. Next, we need to test the robotic fish in several environment configurations after the implementation of an improved motion planner. This includes adding or removing the amount of obstacles in the tank, placing them in different positions, and altering the size of the obstacles. Finally, having a much larger tank in terms of width and length would significantly improve our research as we are currently confined to work in the smaller tank that we have. A larger tank will allow us to test various path-planning and motions of the robotic fish such that it won’t constantly be hitting the walls of the tank.

In the bigger picture, future work should be dedicated towards translation of our system to real-world marine deployment. This will require improvements to system reliability, and an alternative vision system not reliant on ArUco markers. Integration of new sensor hardware such as ultrasonic proximity sensors, depth cameras on the robot, or other optical sensors will be necessary to enable this transition. However, our control and motion planning results provide a promising framework for the future.