Problem

At MIT Lincoln Laboratory's Beaverworks Summer Institute, our team faced the challenge of designing an autonomous underwater vehicle (AUV) capable of operating in high-pressure, deep-water environments. The vehicle needed robust perception and control to navigate complex underwater currents, avoid obstacles, and maintain stable depth and heading during mission-critical tasks. Prior attempts at similar systems were limited by sensor fusion latency and unreliable real-time object tracking, which hindered mission success in harsh conditions.

We required a tight integration of perception, planning, and control to enable autonomous decision-making under pressure. The problem included ensuring reliable detection of nearby AUVs for coordinated operations, as well as maintaining precise depth and heading control using sensor feedback. To validate our approach, we integrated a YOLOv8 model trained for real-time detection, setting a target accuracy that would enable safe, responsive maneuvers in high-stress underwater scenarios. Demo video offered a contextual reference for how real-time detection would inform control decisions.

Solution

We designed and deployed an AUV platform with ROS 2-based software, leveraging Python and C++ for modular perception and control stacks. Our system implemented heading and depth PID controllers, tightly coupled to sensor feedback to achieve stable, responsive navigation in challenging pressure regimes. The perception module integrated sensor fusion to provide a coherent state estimate, enabling robust autonomous decision-making even with noisy measurements.

To address tracking of nearby AUVs, we trained a YOLOv8 model to achieve 90% accuracy in real-time detections, enabling reliable identification and tracking during cooperative missions. We validated performance through simulation and field trials, iterating on calibration, communication protocols, and fault handling to ensure mission resilience in high-pressure environments. The result was an end-to-end solution capable of autonomous operation with predictable control in demanding underwater settings.

Impact

Accuracy improvement: X% detection reliability in real-time operations
Efficiency gain: Y% reduction in mission planning overhead
Reliability metric: Z% of missions completed without major faults

Quote of how incredible of an impact this had.

Team

Team Member #1 - Robotics Engineer
- Led the ROS 2 architecture design and integration

Team Member #2 - Computer Vision Engineer
- Developed and trained YOLOv8 for real-time AUV tracking

AUV Autonomy with Perception and Control