Welcome to the debut of our Operational Deployment Stories. This series pulls back the curtain on the real-world testing and innovative engineering driving the AUTOASSESS project.
Across this series, we will showcase how advanced autonomous drones can enter dangerous, pitch-black, and foggy industrial spaces entirely on their own. You will see these platforms navigate around unexpected hazards, decide what areas require scanning, and interact with massive steel structures to take critical integrity measurements, all without a human pilot in harm’s way.
The Threat: Wind Shear and Structural Hazards
In our first instalment, we highlight the work of our partner UZH (University of Zurich) under Work Package 3 (Resilient Onboard Localisation, Spatial AI, and Control).
When deploying an Inspection Unmanned Aerial System (IUAS) inside confined maritime environments like ballast water tanks or cargo holds, the drone must deal with extreme self-generated airflow and external turbulence. To prove that our drones can accurately track their position relative to their surroundings and dynamically adapt to external environmental threats in real time, UZH developed and field-tested PA-MPPI (Perception-Aware Model Predictive Path Integral Control).
Inside the Technology: What the Videos Show
Our demonstration video shows the real-time closed-loop execution of this controller under rigorous experimental conditions:
1. Real-Time Processing & Onboard Mapping
- What you see: A real-time capture of the drone’s onboard visualization system (Rviz).
- The Science & Technology: The left panels display live camera feeds (including low-light and feature-tracking streams), while the main window shows a sparse 3D point cloud of tracking landmarks (green dots) and the drone’s estimated trajectory (yellow and red lines). This demonstrates the system running tightly coupled visual-inertial odometry entirely on resource-constrained hardware onboard the drone, planning and executing safe paths within milliseconds.
2. Closed-Loop Accuracy vs. Benchmarks
- What you see: A side-by-side comparison of a physical quadrotor navigating a tight testing arena while its 3D trajectory is plotted below.
- The Science & Technology: The live plot benchmark-tests UZH’s custom HDVIO2.0 state estimator against standard commercial hardware (Intel RealSense) and high-fidelity motion-capture Groundtruth. The results show that the UZH framework maintains an tight tracking tolerance, which is critical when a drone is required to fly near boundaries or manholes where a slip of a few centimeters could cause a crash.
3. Disturbances Resilience Testing
- What you see: An autonomous inspection experiment where a quadrotor is tasked with visiting three separate landmarks in sequence. A large industrial fan (highlighted by a red circle) is activated to blast the drone with strong wind disturbances.
- The Science & Technology: Operating on an iterative policy framework, the drone starts with a highly conservative flight path to ensure safety. Over just 2 minutes of runtime (Policy Iteration 12), the algorithm dynamically learns the wind profile, adapts its control outputs, and aggressively counteracts the disturbances. This optimization reduces total travel time by 50% while maintaining absolute path accuracy.
4. Perception-Aware Path Integral Control
- What you see: A technical breakdown of the PA-MPPI architecture operating at 50Hz.
- The Science & Technology: The controller samples dozens of dynamically feasible trajectories within a live 3D occupancy map (ROG-Map). It uses a novel ray-tracing reward system that awards points to trajectories whose endpoints project through unknown space toward the goal, while penalizing paths that intersect obstacles. When compared against state-of-the-art architectures like SUPER, PA-MPPI achieves faster replanning (under 6.33 seconds in highly cluttered settings), proving its superiority for complex structural environments.
What do these videos mean for the maritime sector?
These four clips represent a milestone for the project: we have successfully given these drones a digital “survival instinct”. In the first clip, the drone builds its own map in pitch-black conditions. In the next two, it proves it can fly safely through narrow openings and instantly correct its balance when hit by disturbances. Finally, it uses its onboard processing architecture to safely route itself through cluttered rooms. By proving our platforms can navigate through darkness, blinding camera blur, and unpredictable wind, these videos show that AUTOASSESS has built an autonomous inspection system.
For the maritime sector, the real-time localization and wind resilience shown in these videos represent a practical breakthrough. By proving a drone can map dark, unknown spaces entirely on its own, instantly fight off heavy air currents, and squeeze through narrow gaps, this technology shows that robots can replace human inspectors in dangerous environments. Shipowners can safely deploy automated aerial systems into high-risk ballast tanks and cargo holds, reducing total inspection times, slashing costs, and keeping crews completely out of harm’s way.
What’s next in the Series?
The real-time control loop developed by UZH forms the survival instinct of our robotic fleet, but it is only one piece of the puzzle. In the upcoming instalments of our Operational Deployment Stories, we will explore:
- Semantic 3D Reconstruction: How we built a dense geometric mapping system where a robot scans a hazardous asset to create a flawless 3D model. Human inspectors can simply type a semantic word like “corrosion” or “crack” to instantly highlight every location in the 3D space where structural damage is occurring.
- Cloud-Side Analytics Integration: A look at the front-end visualization interfaces (via Cognite Data Fusion), bringing multi-modal 3D data straight to an inspector’s web browser.
- Full-Scale Industrial Trials: Insights into the deployment of these integrated systems during our exhaustive field-testing campaigns on bulk carriers, container ships, and oil tankers.
Stay tuned as we continue to push the boundaries of what autonomous systems can achieve in ships’ demanding environments!
