Endowing GPT-4 with a Humanoid Body
Building the Bridge Between Off-the-Shelf VLMs and the Physical World
Video Demo
Qualitative Results
Put the stool besides the table and sit on it.
Touch each stack of books on the shelf from top to bottom.
Walk alongside the vehicle.
Walk, jog, crawl and jump.
Sit and touch the plant with your hand.
Walk around the sofa and rest both hands on the backrest.
Strike down one target, and kick down another.
Wake up, walk on stairs to dance on stage.
Comparisons
MoConVQ
- A person is sitting on a sofa and waving hand above head.
BiBo (Ours)
MoConVQ [Yao 2024]
- Strike and kick.
BiBo (Ours)
MoConVQ [Yao 2024]
UniHSI
- A person is leaning against the side of the sofa with hands resting on the armrest.
BiBo (Ours)
UniHSI [Xiao 2024]
- A person is dancing while turning on the light.
BiBo (Ours)
UniHSI [Xiao 2024]
- stand from the sofa.
BiBo (Ours)
UniHSI [Xiao 2024]
HumanVLA
- Carry the box.
BiBo (Ours)
HumanVLA [Xu 2024]
CLoSD
- A person is grabbing a book from the bookshelf.
BiBo (Ours)
CLoSD [Tevet 2024]
2. A person is sitting on the sofa while turning on the lamp on the side table.
BiBo (Ours)
CLoSD [Tevet 2024]
