Paper sets safety rules for sharing human corrections across robots
Original title: When Human Corrections Travel: Safety in Federated VLA Adaptation
Authors want context preserved, closed-loop checks, and leakage measured, but report no experiments.
The paper examines federated sharing of vision-language-action adaptations built from local human corrections, without pooling raw robot records. Corrections vary by modality, safety meaning, and privacy, and a shared update that raises average success may drop a rare safety intervention or act unsafely on another embodiment. The authors argue for three requirements and offer counterexamples plus an evaluation agenda rather than results.
- Sharing spans tasks, environments, people, and robot embodiments.
- Immediate, legible defensive behavior is still required during human interaction.
- The paper presents counterexamples and a falsifiable agenda, not experiments.
From the original article
1 min read at arxiv.orgVision-language-action (VLA) policies can adapt to local human intervention, while federated learning can combine distributed robot experience without pooling raw records. We examine their prospective intersection: sharing adaptations derived from diverse human corrections across tasks, environments, people, and robot embodiments. Corrections differ in modality, safety semantics(urgency, behavioral scope, etc.)
By Vigneshwaran Subbaraju · 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.