Neurosymbolic Architecture for Assembly with Semantic Constraints

A neurosymbolic model learns to assemble novel structures by integrating dense visual observations with symbolic evidence from natural language. Experiments in a simulated truck assembly domain show that using semantic language constraints enables more data-efficient online adaptation than relying solely on task demonstrations.