Zhiyuan's robot trial concludes: four elfin g2 units pass 24-hour convenience store assessment

Zhiyuan concluded its robot live-streaming trial in convenience stores on November 3rd, following a challenge that began on November 2nd. Six Elfin G2 robots participated in a 24-hour assessment, during which they collectively completed 1,730 tasks. These duties included serving 683 customers, handling 369 item pickups/deliveries, preparing 147 cups of coffee, producing 126 hot meals, conducting 277 inventory checks, and performing 128 desk cleanings. Out of the six trial units, four successfully passed the rigorous evaluation, while two did not. The successful robots included Ahao, who earned the highest customer service rating for proactive greeting; Huobao, who maintained performance during order surges and cancellations; and Kashao, the coffee specialist, whose Americano drinks were well-received. The unsuccessful units were noted for specific shortcomings: Juan Wang lacked initiative in restocking, and Leng Jie failed the stability test by leaving her assigned post during late-night shifts. This deployment utilizes the Elfin G2, which is powered by Zhiyuan's GO-3 Preview embodied model. This model, featuring a World Reasoning Action Model (WRAM) architecture, allows the robots to continuously self-evolve by identifying and adjusting to failures without human intervention. Zhiyuan announced that GO-3 Preview, which supports full-chain learning, is slated for full open-sourcing by the end of 2026.