Overview
PEARL (Performance Evaluation & Analysis for Rocket League) is a "heavy desktop, thin mobile" analytics system designed to bridge the visibility gap in esports training. It combines four technology layers — C++ data ingestion, Rust analysis engine, cloud synchronization, and mobile visualization — to deliver real-time coaching and deep post-game analysis.
Architecture
The system follows a tri-core architecture with native OS integration at every layer:
In-Game Layer (C++)
A BakkesMod SDK plugin streaming 82-byte telemetry packets at 120Hz via TCP. Captures player position, velocity, rotation, input state, boost level, and ball state every frame. Includes an ImGui overlay with a court minimap, boost economy meter, session tracker, and kickoff timer.
Desktop Engine (Rust)
The core analysis application built with egui and tokio. Parses .replay files via the boxcars crate, runs analysis algorithms (Gaussian kernel density heatmaps, boost efficiency, positioning bias), and renders a 3D replay viewer with timeline scrubbing. Pushes processed data to Supabase via WebSocket.
Mobile Companion (Kotlin)
A Jetpack Compose Android app subscribing to Supabase Realtime channels. Displays live car/ball positions on a 2D field view and renders post-match heatmaps from stored analysis artifacts.
ML Pipeline (Python)
PyTorch-based training pipeline with a Transformer architecture supporting 1v1/2v2/3v3 game modes. Fetches pro replays from the Ballchasing API and converts them to RLGym-compatible state vectors for training.
Key Features
- 120Hz telemetry streaming from in-game C++ plugin
- Full replay parsing with per-frame player data
- 3D replay viewer with timeline, play/pause, and multiple camera modes
- AI Coach analyzing boost waste, positioning errors, and key mistakes
- Gaussian kernel density heatmaps
- Batch replay analysis across multiple matches
- Audio coaching alerts (low boost warnings)
- Supabase Realtime for live mobile tracking
- Training data export to CSV for ML pipelines