Render Tabs¶

All rendering widgets: Fast Lane live views, Slow Lane telemetry streams, and multi-operator evaluation: are hosted inside RenderTabs, the central QTabWidget at the heart of the MOSAIC UI.

        %%{init: {"flowchart": {"curve": "linear"}} }%%
graph TB
    RT["RenderTabs<br/>(QTabWidget)"]

    subgraph Static["Built-in Tabs (created at startup)"]
        GRID["Grid Tab<br/>(GridRendererStrategy)"]
        RAW["Raw Tab<br/>(text output)"]
        VIDEO["Video Tab<br/>(RgbRendererStrategy)"]
        REPLAY["Human Replay Tab"]
        MULTI["Multi-Operator Tab<br/>(MultiOperatorRenderView)"]
    end

    subgraph Dynamic["Dynamic Tabs (created per-run)"]
        FLT["FastLaneTab<br/>(Qt Quick · ~60 Hz)"]
        LTT["LiveTelemetryTab<br/>(slow lane · ~10 FPS)"]
    end

    RT --> GRID & RAW & VIDEO & REPLAY & MULTI
    RT --> FLT & LTT

    style Static fill:#e8f5e9,stroke:#2e8b57,color:#333
    style Dynamic fill:#e3f2fd,stroke:#1565c0,color:#333

RenderTabs¶

RenderTabs (gym_gui/ui/widgets/render_tabs.py) is a QTabWidget that owns all rendering surfaces.

RenderTabs(
    parent: QWidget | None = None,
    *,
    telemetry_service: TelemetryService | None = None,
    run_manager: TrainingRunManager | None = None,
)

Built-in tabs are created at startup:

Tab	Purpose
Grid	Hosts Rendering Strategies `GridRendererStrategy` and `BoardGameRendererStrategy` with a welcome overlay when idle.
Raw	Plain-text environment output (ASCII renders, debug info).
Video	Hosts `RgbRendererStrategy` for pixel-based environments.
Human Replay	Episode review for previously recorded human sessions.
Multi-Operator	Side-by-side agent comparison via `MultiOperatorRenderView`.

Dynamic tabs are created on demand by worker presenters:

add_dynamic_tab(run_id, name, widget): adds or focuses an agent-specific tab (e.g., CleanRL-Live-agent_0).
Tabs are named by the presenter using the pattern {WorkerName}-Live-{agent_id}.

Signals forwarded from Rendering Strategies board-game renderers:

chess_move_made(str, str)
connect_four_column_clicked(int)
go_intersection_clicked(int, int) / go_pass_requested()
sudoku_cell_selected(int, int) / sudoku_digit_entered(int, int, int) / sudoku_cell_cleared(int, int)
checkers_cell_clicked(int, int)
human_action_submitted(str, int): operator_id, action_index
board_game_move_made(str, str, str): operator_id, from_sq, to_sq
chess_move_button_clicked(str, str): operator_id, uci_move

Worker Presenters¶

Worker presenters (gym_gui/ui/presenters/workers/) are responsible for creating the correct tab type when a training run starts:

Presenter	Tab created
`CleanRlWorkerPresenter`	`FastLaneTab` (live training) + `LiveTelemetryTab` (telemetry)
`XuanCeWorkerPresenter`	`FastLaneTab` with XuanCe-specific naming
`RayWorkerPresenter`	`RayMultiWorkerFastLaneTab` (multi-agent tiled view)
`ChessWorkerPresenter`	`BoardGameFastLaneTab` with move metadata parsing
`HumanWorkerPresenter`	`LiveTelemetryTab` with human input integration

The WorkerPresenterRegistry (registry.py) maps worker IDs to presenter classes so MainWindow can instantiate the correct presenter automatically.

Render Tabs¶

RenderTabs¶

Worker Presenters¶

See Also¶