New GPT-Policy framework enables robots to learn from context

Researchers have introduced GPT-Policy, a novel framework designed to enable robots to learn from context at deployment time. This system integrates a context compiler, a vision-language model (VLM) like GPT-6 Astra to propose actions, and a constrained controller for execution and verification. Experiments show that human video demonstrations improve task completion, with additional gains observed when action references are provided for contact-sensitive tasks. The findings suggest GPT-Policy is a step towards adaptable robots that can translate VLM capabilities into physical actions. AI

IMPACT Enables robots to learn and adapt in real-time, potentially accelerating the deployment of more versatile autonomous systems.

RANK_REASON The item describes a new framework and its evaluation in a research paper submitted to arXiv. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.CV →

• alphaXiv
• arXiv
• CatalyzeX
• CORE Recommender
• DagsHub
• Gotit.pub
• GPT-6 Astra
• GPT-Policy
• Hugging Face
• Influence Flower
• ScienceCast

• model release
• product
• other

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →