Differentiable Parsing and Visual Grounding of Natural Language Instructions for Object Placement
This work addresses object placement tasks for robotics or AI systems, presenting an incremental improvement by combining parsing with probabilistic reasoning.
The paper tackles the problem of grounding natural language instructions for object placement by addressing compositionality and ambiguity, resulting in a fully differentiable method that integrates parsing into a probabilistic learning framework for robust handling of complex inputs.
We present a new method, PARsing And visual GrOuNding (ParaGon), for grounding natural language in object placement tasks. Natural language generally describes objects and spatial relations with compositionality and ambiguity, two major obstacles to effective language grounding. For compositionality, ParaGon parses a language instruction into an object-centric graph representation to ground objects individually. For ambiguity, ParaGon uses a novel particle-based graph neural network to reason about object placements with uncertainty. Essentially, ParaGon integrates a parsing algorithm into a probabilistic, data-driven learning framework. It is fully differentiable and trained end-to-end from data for robustness against complex, ambiguous language input.