Multi-agent Planning using Visual Language Models

ECAI 2024

Pick up the statue from the table and turn on the lamp.(trial_T20190909_193045_208933)

Ground truth Single Agent with table Multi Agent with table Single Agent with image Multi Agent with image
STEP 1: Walk towards the table then turn right and walk forwards.
STEP 2: Turn left and pick up the statue that is on the wooden side-table.
STEP 3: Turn around and walk towards the wooden side-table next to the couch.
STEP 4: Turn on the lamp on top of the table.
Walk to the SideTable at position (-0.589, 0.000, 1.227).,
Pick up the Statue from the SideTable.,
Turn around and walk to the DeskLamp at position (1.657, 0.654, -1.605).,
Turn on the DeskLamp.
Walk to the SideTable at ID 33.,
Check if there is a statue on the SideTable.,
If no statue, walk to the SideTable at ID 37.,
Check if there is a statue on the SideTable.,
If no statue, walk to the SideTable at ID 44.,
Check if there is a statue on the SideTable.,
If no statue, walk to the DiningTable at ID 32.,
Check if there is a statue on the DiningTable.,
Pick up the statue from the table where it is found.,
Turn around and walk to the DeskLamp at ID 23.,
Check if the DeskLamp is operational.,
If DeskLamp is not operational, walk to the FloorLamp at ID 24.,
Check if the FloorLamp is operational.,
If neither lamp is operational, walk to the LightSwitch at ID 25.,
Turn on the LightSwitch.,
Return to the operational lamp and turn it on.
1. Walk to the table in the center of the room.
2. Pick up the statue from the table.
3. Turn to face the lamp on the tripod stand near the couch.
4. Walk to the lamp.
5. Turn on the lamp.
1. Walk to the wooden cabinet on the right side of the room.
2. Pick up the statue from the top of the cabinet.
3. Walk to the floor lamp located on the left side of the room, next to the brown sofa.
4. Locate the switch or knob on the lamp to turn it on.
5. Turn on the lamp by flipping the switch or twisting the knob.
Architecture