Zhiyuan Advances Embodied AI from Demonstration to Deployment with G2 Max Robot
具身智能从“演示”走向“部署”:智元把机器人送进真实工位
At the 2026 Zhiyuan Innovation Day forum in Shanghai, Zhiyuan presented its transition from stage demonstrations to industrial deployment. The company unveiled the GO-3 Preview foundation model, Genie Master agent framework, and the G2 Max heavy-load industrial robot. Specific metrics highlighted improvements in task success rates through collective experience learning and post-training scaling. Early deployments are occurring in electronics manufacturing and retail logistics settings.
Source: Chinese robotics — Company news discovery · Read original article ↗
English translation unavailable. Choose the original article to read it.
What the source reports
Publisher-reported claims, with original evidence. These results have not been independently verified by RoboSignal.
Reported numbers
Reported success rate
97.5%
Summed sensor-hours
View original evidence
Genie Evolver把真机、仿真和世界模型结合起来,提升真实任务中的策略可靠性——手机包装任务中,96台真机组成的群体经验学习系统实现98.3%的任务成功率;手机充电任务中,成功率由60.0%提升至97.5%。
Open source S13Reported success rate
85.7%
Summed sensor-hours
View original evidence
在上海的一家便利店中,搭载Genie Master的精灵G2已在熟食打包、取货、补货等复杂精细任务中持续改进,完成任务流程的成功率从首日的28.6%提升至第五天后稳定在85.7%;在一些难度较大的任务上,部署后的持续改进相比仅依靠监督微调的方式提升17.7%。
Open source S15Reported success rate
28.6%
Summed sensor-hours
View original evidence
在上海的一家便利店中,搭载Genie Master的精灵G2已在熟食打包、取货、补货等复杂精细任务中持续改进,完成任务流程的成功率从首日的28.6%提升至第五天后稳定在85.7%;在一些难度较大的任务上,部署后的持续改进相比仅依靠监督微调的方式提升17.7%。
Open source S15Reported success rate
86.7%
Summed sensor-hours
View original evidence
在插花、叠衣、手机充电、凳子组装、手机包装五项真实任务上,CEL训练的策略全部超过6名人类远程操作员的平均水平,最高反超16.7个百分点——在相同物理训练时间内,参与学习的机器人从16台扩至96台,成功率从86.7%升至98.3%。
Open source S16Reported success rate
98.3%
Summed sensor-hours
View original evidence
在插花、叠衣、手机充电、凳子组装、手机包装五项真实任务上,CEL训练的策略全部超过6名人类远程操作员的平均水平,最高反超16.7个百分点——在相同物理训练时间内,参与学习的机器人从16台扩至96台,成功率从86.7%升至98.3%。
Open source S16Reported success rate
56.4%
Reported zero-shot result; the excerpt does not specify an evaluation protocol.
Summed sensor-hours
View original evidence
现场公布的实验显示:在训练数据规模扩展实验中,模型在20类技能100个任务上的Zero-Shot(零样本)平均成功率由13.4%提升至56.4%;在模型规模扩展实验中,平均成功率由56.2%提升至73.2%,分别展现了数据规模与模型规模扩展带来的能力提升趋势。
Open source S11Reported success rate
13.4%
Reported zero-shot result; the excerpt does not specify an evaluation protocol.
Summed sensor-hours
View original evidence
现场公布的实验显示:在训练数据规模扩展实验中,模型在20类技能100个任务上的Zero-Shot(零样本)平均成功率由13.4%提升至56.4%;在模型规模扩展实验中,平均成功率由56.2%提升至73.2%,分别展现了数据规模与模型规模扩展带来的能力提升趋势。
Open source S11
- paper: not_reported
What remains unknown
Not established in the collected evidence: Environment, Control, Data origin.
Reported performance applies to the described task. It does not establish general autonomy or deployment readiness.
Source excerpts and review record
Automatically extracted; no manual editorial approval recorded.
Genie Evolver把真机、仿真和世界模型结合起来,提升真实任务中的策略可靠性——手机包装任务中,96台真机组成的群体经验学习系统实现98.3%的任务成功率;手机充电任务中,成功率由60.0%提升至97.5%。
Open source S13
在上海的一家便利店中,搭载Genie Master的精灵G2已在熟食打包、取货、补货等复杂精细任务中持续改进,完成任务流程的成功率从首日的28.6%提升至第五天后稳定在85.7%;在一些难度较大的任务上,部署后的持续改进相比仅依靠监督微调的方式提升17.7%。
Open source S15
在插花、叠衣、手机充电、凳子组装、手机包装五项真实任务上,CEL训练的策略全部超过6名人类远程操作员的平均水平,最高反超16.7个百分点——在相同物理训练时间内,参与学习的机器人从16台扩至96台,成功率从86.7%升至98.3%。
Open source S16
现场公布的实验显示:在训练数据规模扩展实验中,模型在20类技能100个任务上的Zero-Shot(零样本)平均成功率由13.4%提升至56.4%;在模型规模扩展实验中,平均成功率由56.2%提升至73.2%,分别展现了数据规模与模型规模扩展带来的能力提升趋势。
Open source S11
Source:Chinese robotics — Company news discovery · msn.cn