Repository navigation
Enhanced Model Matching System
Note
This functionality was removed from version 26.2 onward, due to the costs of model enhancing.
Model matching across different flight simulators has always been one of the major challenges for JoinFS, the multiplayer application enabling pilots from different simulators to fly together. Due to inconsistent naming conventions and highly varied third-party add-ons, correctly identifying and matching a remote aircraft to a local equivalent has historically required manual rule sets and guesswork.
The new Enhanced Model Matching System introduces an AI-assisted, vector-driven workflow that improves accuracy and robustness across heterogeneous simulators.
JoinFS begins by scanning the local simulator and retrieving the available aircraft models and liveries, including:
- Aircraft title
- Livery name
This raw data is used as the basis for the enrichment process.
To keep the JoinFS client lightweight, an external web service performs the enrichment:
- Protects AI provider API keys
- Implements heavy caching to avoid repeated AI calls
- Ensures stable and predictable throughput
- Provides a consistent, simulator-independent output format
The web service uses the Gemini AI model to enhance and normalize the aircraft data. The enriched output includes structured semantic features that are later used for similarity-based matching:
Enriched Feature Set:
- Role: general aviation, airliner, glider, helicopter, etc.
- Wing Position: high-wing, low-wing, mid-wing, etc.
- Number of Engines: single-engine, twin-engine, quad-jet, etc.
- Engine Type: piston, turboprop, turbofan, turbojet, electric, etc.
- Wake Turbulence Category: light, medium, heavy, super
- Military or Civilian: distinguishes military aircraft from civil operators
These semantics allow vector-based algorithms to compare aircraft in a far more meaningful way than raw text titles.
The enriched text and structured features are combined into a single descriptive representation. This description is then passed through a tokenizer to produce a multi-dimensional vector embedding.
Benefits of vectorization:
- Encodes aircraft semantics rather than literal names
- Allows precise mathematical similarity comparisons
- Works even with highly inconsistent naming schemes
- Enables robust cross-simulator matching
When another pilot connects:
- The remote aircraft’s model/livery undergoes the same enrichment and vectorization pipeline.
- JoinFS compares the remote vector against all local aircraft vectors using similarity metrics (cosine similarity).
- The aircraft with the highest similarity score is selected as the best match.
This results in accurate matches even across entirely different simulator ecosystems.
The subjective benefits of the enhanced model matching are:
- Far more accurate model matching across multiple simulators
- Semantic understanding of aircraft types and roles
- Reduced need for manual matching files
- Resilience to inconsistent or incomplete naming
- Caching reduces API usage and increases performance
- Future-proof as new aircraft add-ons are released
The performance-toll of the model matching is split into these categories (the figures are measured on my mid-range computer):
- Initial enrichment of the users available models (0.07s for 1215 unique models in my simulator)
- Initial computation of the vector embedding (9.16-12.05s for 2335 models). This initial computation is deferred to a background thread, the user can use JoinFS without being blocked by these computations.
- Enrichment of every model that is read from the network users (within the error margin of the .net stopwatch)
- Embedding computation of the network users models (0.007-0.015s per model)
- Vector comparison between the users and network model for the final selection of the model presented to the user (within the error margin of the .net stopwatch)
Should you wish to opt-out of the AI-enchanced model matching, uncheck the new settings "Use AI Model Matching". By default this setting is checked, meaning that the enhanced model matching is used by default.
The new Enhanced Model Matching System represents an improvement in how JoinFS interprets and matches aircraft models across heterogeneous flight simulator ecosystems. By combining AI-driven enrichment with vector-based similarity search, model matching becomes more accurate, consistent, and resilient.