Our Technology - Orca AI

Data collection

Fully Automated Data Pipeline

Our unique methodology collects only data crucial for training our AI algorithms.
Our algorithms have prioritized, filtered, and synchronized five years’ worth of sailing data from over 200+ ships.
Only 5% of the entire data is relevant for training in terms of maritime-sensitive operations.
The data pipeline is fully automated with no manual interventions required.
The platform fuses raw data from navigation sensors, such as GPS, Radar S/X band, AIS, Water depth, Wind, Rate-Of-Turn indicator, and Gyrocompass.

Data labelling

All raw data is comprehensively labeled using our proprietary automated labeling system to ensure the highest accuracy.
Orca AI’s automated labeling algorithms boost human productivity by a factor of 10.
All data relevant for the scenario is labeled: visibility conditions, congestion level, shore, clouds, and marine targets.
Detailed evaluation of the quality of the AI algorithms under all such conditions further improves the performance of the AI.

Evaluation

We constantly evaluate the platform’s algorithmic predictions versus the real world, providing optimal metrics for navigational efficiency.
Our tracking technology boosts performance using the ‘majority voting’ machine learning technique to increase prediction accuracy.
Tracker and filters achieve 99.9% accuracy and are able to identify and filter out false positives.
Ongoing assessments are carried out and approved every time an updated AI model is distributed to our customer fleet.

The enhanced object identification capability of our 6th generation detector can be seen in the comparison below:

5th Generation Detector

6th Generation Detector

Deployment

Each ‘ever-ready’ new model is deployed in our simulation lab, where we inject real-world data and review performance at scale.
Once the artificial intelligence’s performance exceeds the previous model, it is deployed on selected ships where it is tested in live operations.
The vessel masters and crew provide valuable feedback.
Our monitoring platform tracks system performance in order to fine-tune the AI model as necessary.
The improved AI algorithm is distributed to our entire customer fleet; this happens every time we upgrade our AI.