Periodic Reporting for period 2 - SmartRunway (AI boosting airport capacity)
Période du rapport: 2021-06-01 au 2023-04-30
Currently, air traffic control (ATC) is a highly manual process that relies on Air Traffic Controllers using equipment with minimal automation. Despite decades of research and expectations of automated solutions, ATCOs still rely on visual supervision of positions on a radar screen with limited alerts. This leads to a non-scalable system, facing a shortage of qualified personnel to meet expected traffic growth causing capacity constraints, delays, and environmental inefficiencies.
Our solution, Victor5, is a disruptive technology that brings machine learning to ATC. Victor5 offers a range of use cases that increase operational efficiency in the air traffic control room. It is a full-stack platform that enables the creation of a data lake for any airspace operation, allowing for in-depth analysis of environmental efficiency, safety assessments, workload, and providing a platform for higher levels of automation powered by AI. With Victor5, we aim to revolutionize the way ATC is conducted, providing a trustworthy and innovative solution to meet the growing demands of the aviation industry.
The first ready-to-deploy use case for Victor5 is a contingency and situation awareness service for air traffic control operations in the event of service disruption that requires transitioning to ATC Zero or maintaining operations in a degraded mode. In general, Victor5 addresses contingency situations where Air Traffic Control (ATC) sub-systems fail, for en route, tower, or approach. On occasion, the situation requires that operational personnel vacate the ACC or TWR building and move to a secondary facility that may or may not have additional backup systems. Victor5 allows operational personnel to maintain situational awareness during such transitions to backup equipment but also to maintain degraded operations (if the backup equipment does not exist or is not available due to a common failure with the primary equipment, like radar feed) or to safely clear the skies.
Victor5 assists the operational room with a portable device that meets the following requirements:
- A reliable ATC data feed(s) with real-time quality assurance.
- A portable device with an ATC UI displaying ATC console information, adapted for emergency scenarios in mind.
- Digital tools to ease the controller’s workload in stressful situations, including short-term conflict alert (STCA) and medium-term conflict detection (MTCD).
Victor5's computational power enables the development of various use cases, from batch training to real-time data processing, all utilizing the same data structures. Some of the current use cases include:
- En-route planner air traffic controller digital assistant: Victor5 can assist planner controllers by performing tens of thousands of distance comparisons and running a fuel model in real-time to identify the best course of action for the executive controller.
- Supervisor digital assistant for sector configuration and workload: Victor5 performs up to 50,000 distance comparisons per minute to estimate traffic demand and workload for air traffic controllers. Using real-time wind information and advanced ML algorithms, supervisors have a simple interface with a prediction that encapsulates the computational power of the platform.
- Post-operations safety and environmental analysis: Victor5's data lake is enriched with information about operations in any en-route sector, approach, or runway/taxiway/apron. This allows to inform safety analysts, managers, and chiefs of operations about safety performance, fuel consumption, and ATC system performance.
Technically, we have improved the state of the art in three areas:
1 Data architectures - We have developed a data architecture running in the cloud capable of processing real time a virtually unlimited number of traffic positions without delay. Our multi-environment cloud infrastructure is based on micro-services deployed on containers. This allows us to scale our software components in a flexible, resilient and automated way through clusters. We rely on clusters processing to read incoming data coming from our data lakehouse and the input is processed by our data pipelines, delivering the analysis outcomes to our machine applications and frontend layer.
2 Level5, a set of AI algorithms for ATC trained with large historical datasets, including positions and flight plans. Level5 provides real time key predictions, labelling conflicting and non-conflicting traffic, identifying aircraft that need an instruction to manage conflicts. We build on machine learning pipelines for scale, efficiency, and control. For model building and training we rely on state-of-the-art libraries and every experiment and model deployment is traced, which allows us to use a fine model selection and standardized offline validation procedure. Finally, we rely on a complete observability platform that monitors our infrastructure, providing all the insights necessary for online validation and testing.
3 Victor5 UI, a revolutionary interface running on a portable digital assistant for the operational control room, capable of meeting the most strict cybersecurity requirements, running on-premise or in the cloud, through a regular internet or a private 5G connection. The device complements the existing ATC consoles as a contingency tool in case of service disruption, for which no complete solution exists today. We deliver Victor5 with the latest web technologies.