Execution-Aligned Progressive Noise for Consistent Asynchronous Replanning in Generative Robot Policies
EAPN aligns shared noise with executed motion so generative robot policies replan without inconsistent mode switches.
Execution-Aligned Progressive Noise (EAPN) adds structured stochasticity so generative robot policies can replan asynchronously without restarting from independent noise. A shared noise trajectory is aligned with executed displacement between chunks, while temporal correlation is modeled inside each action chunk and combined with committed action context. On D3IL it improves multimodal consistency, and it reports 88.59% average success on Kinetix. Real-robot runs reach 90.0% on Object Storage and 96.7% on bimanual Cloth Folding, with strong LIBERO performance even under long inference delays.
- EAPN shares a noise trajectory aligned with actual execution displacement across chunks.
- Intra-chunk noise models temporal correlation along action time.
- Average success is 88.59% on Kinetix and stays robust on LIBERO under long inference delays.
- Real-robot success reaches 90.0% on Object Storage and 96.7% on bimanual Cloth Folding.
From the original article
1 min read at huggingface.co
Continuous asynchronous replanning is essential for real-time generative robot policies, but independent stochastic initialization can cause mode switching and inconsistent continuation across action chunks. We propose Execution-Aligned Progressive Noise (EAPN), which introduces structured stochasticity at both inter-chunk and intra-chunk levels.
By Di Wu, Ping Liu, Xuhua Chen · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.