With the TeamTime session on February 13 and the task ideas for RoboSub 2025 released, our team now has a much clearer understanding of this year’s competition challenges. Many of the tasks, such as the coin flip, gate, style, bins, torpedoes, and octagon, remain largely unchanged from last year. However, there are two new additions: a slalom-style navigation task that follows the gate challenge and extra points to be gained by returning to the gate at the end of the run.
Strategic Adjustments Based on Last Year’s Performance
Last year, our main robot, Oogway, struggled with inaccurate IMU readings, which caused significant disorientation after completing the barrel roll task and impaired navigation for the rest of the run. To address this issue, we are leveraging our new mini-robot, Crush, which is equipped with a DVL, IMU, and mono cameras. We have divided the tasks strategically between the two robots:
Crush’s Task Flow
Crush will begin its run with the coin flip task before using computer vision (CV) to detect the reef shark and sawfish markers on the gate and navigate through it. After completing the barrel roll for style points, Crush will use CV to align itself with the entrance of the slalom task, ensuring it is centered between the first set of PVC pipes. The slalom maneuver will be executed using either CV-based tracking or dead reckoning, depending on which approach proves more reliable during testing.
Oogway’s Task Flow
With Crush handling the movement-based tasks, Oogway is free to prioritize more complicated tasks that are further away from the start gate such as bins and torpedoes. Oogway has recently been upgraded with a new torpedo system capable of firing at distances up to 2 meters. Additionally, we still have our marker dropper system from last year, along with a well-tested algorithm for the bins task. We hence plan to attempt both of these tasks this year with Oogway. Once these tasks are completed, Oogway will navigate to the octagon task and surface inside the octagon. Finally, both robots will navigate back to the start gate to secure extra points for the final task.
Challenges in Navigation & Object Detection
Since Oogway will no longer follow the slalom course, we need an alternative method for navigation between tasks, specifically to locate the bins and torpedoes reliably. Our acoustics-based navigation algorithm is still under development, so as a fallback, we will rely on CV-based navigation:
- Torpedo Task: The torpedo buoy is relatively easy to detect due to the large upright buoy with the reef shark and sawfish markers, which our DepthAI model can reliably recognize from any point in the course, even the start gate.
- Bins Task: Detecting the bins is more challenging since their images are on the bottom surface and are only visible when the robot is directly above them. To work around that, we are training a CV model to identify the white side of the bins from a distance, which allows us to navigate towards them more effectively. Our next step is to train and evaluate the performance of this model, which will help us determine if this approach is feasible for consistent detection in competition conditions.
Inter-Vehicle Communication (IVC) Strategy
With two eligible AUVs competing this year, one of our most ambitious improvements is implementing IVC to coordinate Oogway and Crush for up to 1000 extra points. We are currently developing our acoustic-based IVC system. This will enable the two robots to communicate and log critical events at key milestones during the run, which will be recorded and presented to the judges as proof of successful communication. As testing progresses, we will refine this approach to maximize its reliability.