Reasoning Table Placement

reasoning_table_placement.py

Overview

reasoning_table_placement.py is a bridge between cognition and the manipulation stack. It listens to high-level decisions on /{robot}/placement_decision and maps them onto low-level commands that the arm / gripper planners understand.

Interfaces (strongly-typed, stateless)

Direction	Topic	Message type	Notes
Required	/{robot}/placement_decision	std_msgs/String	e.g. "Decision: PLACE, IGNORE"
Provided	/{robot}/table_reasoning_commands	std_msgs/String	"PLACE_DISH" | "CLEAR_TABLE" | "NO_ACTION"

Food Placement Integration KPIs

Metric	Target
Placement decision time (/placement_decision → /table_reasoning_commands)	≤ 200 ms
Spot-detection accuracy (correctly free & stable spots)	≥ 95 %
Replanning responsiveness (after clutter update)	≤ 100 ms

Decision Mapping

Decision	Published cmd	Rationale
PLACE	PLACE_DISH	Room for one more plate
CLEAR	CLEAR_TABLE	Table finished; remove crockery
other	NO_ACTION	Either “FULL” or “IGNORE” → do nothing

Contract

Pre-conditions

• Upstream decision topic must contain “Decision:” and one keyword.

Post-conditions

• Exactly one command published per decision message.

• Mapper never blocks; if queue is full older commands drop first (queue_size=10).

Implementation notes

• Uses simple substring checks so future metadata (timestamps, scores) won’t break parsing.

• Publisher is latched at 10 messages to absorb bursts.

• All log lines throttled or filtered to keep console tidy.

class reasoning_table_placement.ReasoningTablePlacement

Bases: object

Runtime object that wires subscriber → callback → publisher.

1. Initialise ROS under {robot}_reasoning_table_placement_node

2. Subscribe to high-level decisions.

3. Advertise low-level commands.

decision_callback(msg)

Translate Decision → Command.

Translation logic

• contains PLACE → PLACE_DISH

• else contains CLEAR → CLEAR_TABLE

• otherwise → NO_ACTION

reasoning_table_placement.main()

Instantiate node and hand control to ROS spin loop.