Maintenance patrol

Connect perception, memory, action selection and route following.

A robot patrols a station, detects a fault and moves to repair it. A door may close between it and the target. Store the robot’s known target position separately from the actual world state.

A small scene or step-by-step model with an event log

• Use one map, one robot and one fault. Specify the graph and edge costs explicitly.
• The robot learns through observations only; its memory expires.
• Every action change logs time, observation and reason.

1. Define patrol, move-to-fault, repair and search states, including transition conditions.
2. Separate observations from memory. Losing sight must not instantly reveal the target’s new position.
3. Connect action selection to path search and movement; recompute after a closing door or return an explicit failure.
4. Repeat these scenarios: visible target, lost target, closing door and unreachable target. Preserve identical starting conditions.

A runnable scene or reproducible step-by-step model, map, transition table and four logs explaining decisions.

• Identical starting conditions and observations produce identical decisions.
• After losing sight, the robot uses the last known position until memory expires.
• A closed door causes neither wall traversal nor endless path recomputation each frame.
• Target reached, no route and start equal to goal are distinguishable results.

Which wrong decision came from memory, which from priorities and which from navigation? Change only one of these layers.