CARP: An Economical Underwater Robot for Algorithm Testing

Abstract

This project discusses the development of a low-cost AUV called the CARP for the Leonard Lab at Princeton University. The end goal of the project is to build a reliable, highly controllable robot for testing multi-agent algorithms. The work done by others over the summer of 2016 and 2017 was taken forward in three ways. First, the external and internal hardware was upgraded: the size of the robot was increased to fit a powerful microprocessor, custom PCB and on-board camera. Existing literature, CAD software and experimentation techniques were used to calculate unknown coefficients as accurately as possible. Second, an LQR controller with multi-rate Kalman filtering was designed. AprilTags, a depth sensor and an IMU are used to measure the CARP’s location and orientation in the tank. Through simulations, the designed controller is shown to robustly control the CARP along 6-DOF. A path-planning algorithm was designed, but not implemented, as it made the performance of the controller worse. Extensive experimental testing could not be conducted due to unforeseen hardware and software issues that hindered progress. Experimental results from testing control along the x -axis are presented. Finally, a robust software architecture was built on ROS, allowing both Real World and Simulation modes as well as seamless transitions between the modes. The ROS platform built for the CARP is robust in that it can be used as a framework for controlling not only other underwater robots, but other robotic vehicles as well. In future works, more extensive experimental testing is required. Once the CARP can be controlled along 6-DOF, it can be used as an agent for physical implementation of theory developed in the Leonard Lab.