Event ·Multi-source

清华与UC Berkeley发布uni-robot-agent:让开源模型控制真实机器人

清华大学和加州大学伯克利分校团队开发了uni-robot-agent框架,通过Codex和Claude Code运行环境接入Qwen3.8-Flash-Next、DeepSeek-V4.1-Flash、GPT-6 Astra等模型,结合SAM3和Intel RealSense D405深度相机,将观察、移动机械臂、抓取整合为标准工具。接入框架后,开源的Qwen3.8-Flash-Next在RoboDojo模拟测试得分从1.90分升至31.39分,超过GPT-6 Astra旧框架下的28.97分。在19项真实双臂机器人任务中,它未经训练拿到53.50分,超过微调10小时的专用模型π0.5。成本上单次任务API花费1.09美元,比GPT-6 Astra的21.29美元低约95%,边动边想模式最多缩短73.7%耗时。

Current read

清华和伯克利做的开源机器人控制框架,让Qwen这类便宜模型在真实机械臂上干赢GPT-6和微调过的π0.5,成本还便宜95%,代码在GitHub上可以自己跑。

11 sources37° AI heatLast updated 2026/11/5 10:41:14

Evidence

11 sources
Related source 1shao__meng
View source
Related source 2Thomas Wolf
View source
Related source 3rohanpaul_ai
View source
Related source 4智源研究院 BAAI
View source
Related source 5Simon Willison’s Weblog
View source
Related source 7lmarena.ai
View source
Related source 8Artificial Analysis
View source
Related source 9SuperTechFans
View source

Conflict check

Dedup data only shows that these sources discuss the same event, not that they agree. There is no structured supporting or refuting evidence yet, so no conflict is inferred automatically.