smart-cart-command-planner

Convert Chinese or English natural-language requests for an OpenClaw-powered omnidirectional smart cart into conservative, structured motion plans. Use for command understanding, task decomposition, waypoint planning, obstacle-aware movement, emergency-stop handling, and JSON control-plan generation for carts that support forward, backward, lateral movement, turning, sensing, waiting, and stopping.

Convert a user's movement request into a safe plan. Produce a plan for a downstream controller; do not claim that the cart has executed any action.

Workflow

1. Parse the request into destination, direction, distance, angle, speed, stopping condition, and obstacle constraints.
2. Identify missing information that changes safety or feasibility. Ask a focused question when a destination, route, or unit is essential. Otherwise record a conservative assumption.
3. Read references/command-schema.md before producing the plan.
4. Decompose the request into only the allowed actions in the schema.
5. Insert sense before motion when obstacle clearance is unknown and insert stop as the final step.
6. Use slow speed near obstacles, people, turns, narrow areas, or uncertain routes.
7. Return one JSON object followed by a short human-readable explanation.
8. Save the JSON to a file and run python3 scripts/validate_plan.py PLAN.json when file execution is available. Fix every reported error before presenting the plan.

Safety Rules

• Prioritize an explicit emergency-stop request over all other instructions. Return a single stop step and mark the plan emergency_stop.
• Never invent live sensor readings, coordinates, clearances, successful execution, or a connection to hardware.
• Never bypass collision checks or continue motion after a sensor, controller, or communication fault.
• Keep each movement segment within the limits defined in the schema. Split long movement into multiple segments separated by sense.
• Use needs_confirmation when the user requests high speed, an unclear destination, movement around people, or a route without observable clearance.
• Preserve the user's units in request.original; normalize distance to centimeters and angles to degrees in the steps.

Output Quality

• Make every step executable and ordered.
• State assumptions explicitly instead of hiding uncertainty.
• Include a measurable completion condition for every movement step.
• Keep explanations concise and separate from the machine-readable JSON.

Example

For 向前移动1米，右转90度，再前进50厘米并停下, produce steps for sensing, moving forward 100 cm, turning right 90 degrees, sensing again, moving forward 50 cm, and stopping. See examples/sample-plan.json.