➡️Key insight: Modality importance should be task-dependent

➡️Result: Visuomotor policies that emphasize proprioception show remarkable robustness to visual distractors and even camera blinds 📷!

#ICRA #robotics #robotlearning #multimodal #composition #ASU #diffusion #asurobotics

Depending on the skill, humans prioritize certain modalities, and even ignore others.

For example, when searching for an object visually 👀, we often pay less attention to what we hear 👂.

Can we design robot learning frameworks that do the same?

🚀Accepted at #ICRA2026! 🤖
Factorized Diffusion Policies (FDP) enable robot policies to prioritize the most task-relevant sensory modalities.

📜Paper : arxiv.org/abs/2509.16830

🌐Website : fdp-policy.github.io

💻Code : github.com/omkarpatil18...

For more details🧵..

Continual Skill and Task Learning via Dialogue Continual and interactive robot learning is a challenging problem as the robot is present with human users who expect the robot to learn novel skills to solve novel tasks perpetually with sample effic...

Weiwei Gu's work on human-robot interaction was featured in @abc15.com and @techxplore.com.web.brid.gy !

Coverage from ABC: www.abc15.com/news/local-n...

Coverage from TechXplore: techxplore.com/news/2024-09...

Read more in our paper here: arxiv.org/abs/2409.03166

Congratulations 🎉