Zhiyuan (AgiBot) launches Genie Master, a self-improving framework for G2 robots
Overview
Zhiyuan, also known as AgiBot, has launched Genie Master, an embodied agent framework that lets its G2 robots improve after deployment instead of waiting for engineers to retrain them.
QbitAI reports that the robots diagnose failures and adjust strategies in a real convenience store without human intervention, and that a 24-hour live test with six G2 robots is planned for November 2.
Pandaily reports that the company says the robots are working in the store on ready-meal packing, picking and shelf restocking, and fixing their own failure modes without human demonstrations. AgiBot also says the second improvement track still uses human takeover data and model training, so people are not removed entirely.
The framework has two tracks: Harness RSI updates prompts, memory and execution rules without changing model weights, and Policy RSI retrains the action policy from execution data.
Written by AI from the articles below · updated Oct 27, 10:28 PM ET
Check the sources:
Article timeline
The articles in this story. Times are ET.
- PandailyNewsAgiBot launches Genie Master, a framework for self-improving G2 robots
AIAgiBot has released Genie Master, an embodied agent framework that lets G2 robots improve after deployment. The company says robots running it are working in a real convenience store, diagnosing and fixing their own failures without human demonstrations. Improvement runs on two tracks: Harness RSI refines prompts, memory and execution rules without changing model weights, while Policy RSI retrains the action policy using data that still includes human takeovers.
- QbitAINewsZhiyuan launches Genie Master, a framework for self-improving convenience-store robots
AIZhiyuan has launched Genie Master, an embodied agent framework that lets its Genie G2 robots diagnose failures and adjust strategies in a real convenience store without human intervention. The framework uses an Embodied Execution Protocol to link the agent's reasoning with robot actions, and combines Harness RSI, which stores experience as memory and rules, with Policy RSI, which updates the action policy from execution data. Zhiyuan plans a 24-hour live convenience-store test with six G2 robots on November 2.
Heat Trend
- Current comparable heat
- 13
- Peak comparable heat
- 16Oct 27, 10:00 PM ET
- Change in past 24 hours
- –
The trend compares only continuously observed participants, so its scope may be smaller than the current heat total. Hover or tap for hourly heat; use arrow keys with a keyboard.