UNCOM integrates off-the-shelf multimodal AI models into a modular zero-shot system that parses commands into object-action-target representations and achieves 82.39% success on a real-world tabletop HRI benchmark with public code and data.
Learning actions from natural language instructions using an on-world embodied cognitive architecture,
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.RO 1years
2024 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
UNCOM: Zero-shot Context-Aware Command Understanding for Tabletop Scenarios
UNCOM integrates off-the-shelf multimodal AI models into a modular zero-shot system that parses commands into object-action-target representations and achieves 82.39% success on a real-world tabletop HRI benchmark with public code and data.