ODYSS:AI - Full Autonomy pour la dépollution des océans

Autonomous detection and collection of marine debris require ensuring the reliability of every link in the chain, from perception to decision-making. To meet this challenge, our team has designed and tested a fully autonomous system capable of perceiving its environment, building a tactical picture, and planning or adapting its missions in real time to navigate safely.

context

The project aims to automate the detection and recovery of floating waste (bottles, floating debris, oil slicks, etc.) using a maritime drone, in partnership with a company specializing in ocean cleanup.

Reducing the need for manual supervision is a key lever for scaling up, which must not come at the expense of safety or reliability. To enable the algorithm to correctly perceive its environment, detect obstacles and plan its movements autonomously, our team developed several technological building blocks: sensor data fusion, the construction of a coherent tactical picture, mission planning and its adaptation to local hazards, and the management of interactions with a human operator, who supervises operations and can validate the collection of detected waste in cases of uncertainty.

Before any deployment on the boat, it was necessary to be able to test, simulate and replay all of these behaviors in a controlled environment, faithfully reproducing navigation conditions (waves, currents, sensor noise) as well as the other entities present around the boat.

innovations

To transform a collection of disparate technology building blocks into a robust, self-contained system, our team implemented several innovations spanning simulation, testing, and industrialization.

On-Board Dynamics Simulator

Development of a simulator that replicates the boat's dynamics and generates realistic sensor data (motion, noise, AIS signals from surrounding vessels), including environmental forces such as waves and currents.

Tactical Situation and Data Integration

Construction of a coherent tactical picture based on simulated or real sensor data, enabling the system to detect obstacles and surrounding objects (including debris) in real time.

Planification de mission adaptative

Développement d'une planification otpimisée des missions, et de la capacité de générer des manœuvres d'évitement d'obstacles puis de reconstruire une trajectoire de retour vers la mission initiale.

Test and Replay Architecture

Implementation of a comprehensive test bench that allows users to view behavior in real time, simulate data, or replay past real-world scenarios for analysis or debugging.

Industrialization of Algorithmic Building Blocks

Consolidation of the various algorithms developed separately into a single architecture, ready to be deployed and tested on the actual drone.

SOLUTION

Develop and integrate innovative algorithmic building blocks into a fully autonomous system that is ready for field use.

This architecture now makes it possible to perceive, decide, and act in real time, while still allowing certain critical decisions to be validated by an operator.

benefits

Thanks to this architecture, our partner now has a system that can operate autonomously while remaining controllable. The concrete benefits are as follows: