Modular Networks for Compositional Instruction Following
This addresses a key limitation in AI systems for tasks like robotics or virtual assistants, though it appears incremental as it builds on existing modular and compositional methods.
The paper tackles the problem of instruction following architectures struggling with novel compositions of subgoals by proposing a modular architecture that segments instructions and uses specialized modules for each subgoal type. It shows improved generalization to novel subgoal compositions and unseen environments on the ALFRED benchmark compared to standard non-modular approaches.
Standard architectures used in instruction following often struggle on novel compositions of subgoals (e.g. navigating to landmarks or picking up objects) observed during training. We propose a modular architecture for following natural language instructions that describe sequences of diverse subgoals. In our approach, subgoal modules each carry out natural language instructions for a specific subgoal type. A sequence of modules to execute is chosen by learning to segment the instructions and predicting a subgoal type for each segment. When compared to standard, non-modular sequence-to-sequence approaches on ALFRED, a challenging instruction following benchmark, we find that modularization improves generalization to novel subgoal compositions, as well as to environments unseen in training.