Multi-Modal Manipulation via Multi-Modal Policy Consensus

Effectively integrating diverse sensory representations is crucial for robust robotic manipulation. However, the typical approach of feature concatenation is often suboptimal: dominant modalities such as vision can overwhelm sparse but critical signals like touch in contact-rich tasks, and monolithic architectures cannot flexibly incorporate new or missing modalities without retraining. Our method factorizes the policy into a set of diffusion models, each specialized for a single representation (e.g., vision or touch), and employs a router network that learns consensus weights to adaptively combine their contributions, enabling incremental integration of new representations. We evaluate our approach on real-world tasks such as occluded object picking, in-hand spoon reorientation, and puzzle insertion, as well as simulated manipulation tasks in RLBench, where it significantly outperforms feature-concatenation baselines on scenarios requiring multimodal reasoning. Our policy further demonstrates robustness to physical perturbations and sensor corruption. We further conduct perturbation-based importance analysis, which reveals adaptive shifts between modalities, for example transitioning from vision to multi-model when entering occluded spaces.