Hierarchical Residual Policy Learning for Real-World Mobile Manipulation with Sparse Human Guidance
HiRe-MoMa fine-tunes real-world mobile manipulation policies in about 30 minutes using sparse human guidance.
HiRe-MoMa is a hierarchical residual framework for sample-efficient real-world fine-tuning of mobile manipulation policies. A frozen base policy receives additive corrections from separate mobile-base and arm residual heads with unidirectional conditioning, reducing gradient interference. Expert-Guided Value Alignment uses sparse human interventions covering under 20% of training to reshape the critic. On the TIAGo++ platform, the method improved task performance after about 30 minutes of online interaction across different base policy architectures.
- HiRe-MoMa fine-tunes a frozen base policy with additive residual corrections.
- Separate residual heads control the mobile base and the manipulator arm.
- Sparse human interventions cover under 20% of training and reshape the critic.
- Real-world TIAGo++ fine-tuning used about 30 minutes of online interaction.
- Products
- TIAGo++
From the original article
1 min read at arxiv.orgMobile Manipulation (MoMa) is a cornerstone of autonomous robotic systems, yet real-world fine-tuning remains a formidable challenge because of high-dimensional action spaces and intricate coupling between heterogeneous modules. In this paper, we propose HiRe-MoMa (Hierarchical Residual Mobile Manipulation), a systematic framework for stable and sample-efficient real-world fine-tuning of complex MoMa systems.
By Zhefei Gong · arxiv.org
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.