MorphAct generates context-dependent updates for robot policies
Original title: MorphAct: Robot Policies Morphing with the Worlds They Act In
The paper reports gains of up to 16.30% on LIBERO and Meta-World and up to 14% over π0.5 in real-world tests.
MorphAct is a contextual parameterization framework that changes robot policy parameters as interaction contexts change, rather than keeping shared parameters fixed. Coupled generators produce low-rank updates for vision-language and action modules and are trained jointly on the native action objective without operation labels. The paper reports improvements across three VLA backbones on LIBERO and Meta-World, plus real-world gains over π0.5, with ablations supporting contextual generation and representation coupling.
- Diagnostics found competing update directions across tasks.
- Vision-language updates condition action-side parameter generation.
- Training uses no operation labels or expert-adapter targets.
- Reported benchmark gains reach 16.30%.
From the original article
1 min read at huggingface.co
Progress toward general intelligence in the physical world requires scaling robot learning across diverse environments and interactions. However, existing robot policies use shared parameters that remain fixed across interaction contexts. When different operations favor competing parameter adjustments, fitting their mixture can leave individual operations insufficiently adapted.
By Ziyan Liu, Jiting Liu, Weihong Zhong · 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.