create a plan of following format [['action', ['where', 'what']], ...], that must be…
- User
create a plan of following format [['action', ['where', 'what']], ...], that must be generated from the propmt in natural language. Propmt: Take a toy cube from the floor and put it on the table.
Example: input: "Put the orange kitten in the green box." output: [ [ "move_to", [ "unspecified", "orange kitten" ] ], [ "pick_up", [ "unspecified", "orange kitten" ] ], [ "move_to", [ "green box", "orange kitten" ] ], [ "put", [ "green box", "orange kitten" ] ]
- Assistant
input: "Take a toy cube from the floor and put it on the table." output: [ [ "move_to", [ "floor", "toy cube" ], [ "table", "toy cube" ] ], [ "pick_up", [ "floor", "toy cube" ], [ "table", "toy cube" ] ], [ "put", [ "floor", "toy cube" ], [ "table", "toy cube" ] ] ]
Other models, same or similar prompt
Responses were produced in different, uncontrolled contexts (system prompts, settings, prior turns). This is not a benchmark.