Perception and Memory

This lane keeps the first robot-facing examples small: synthetic detections, simple belief updates, and typed outputs that are easy to inspect.

Detection flowpixi run -e golden-local demo-perception-detection-flow

Belief flowpixi run -e golden-local demo-memory-belief-flow

Point with memorypixi run -e golden-local demo-memory-pointing-flow

What To Watch

Step	What changes	Retriever concept
Perception	Synthetic frames become typed detections.	Flow I/O and typed payloads.
Memory	Detections update local belief state.	Local Flow state and replayable inputs.
Pointing	A goal is resolved using current belief.	Closed-loop handoff between perception, memory, and action selection.

Use this lane before real cameras or model-backed perception. These commands are smoke checks: success is exit code 0 with the Pixi task line, not a long terminal report. If a command fails, the Flow, task, docs page, and expected output are out of sync.

Expected Results

Command	Stable success cue
pixi run -e golden-local demo-perception-detection-flow	exits 0 after stepping the synthetic camera -> detector -> printer graph.
pixi run -e golden-local demo-memory-belief-flow	exits 0 after stepping synthetic detections through the belief tracker.
pixi run -e golden-local demo-memory-pointing-flow	exits 0 after dropout and belief memory preserve a target point through intermittent detections.