Qingyan Technology Joins Consortium to Release ReAct-Manip Benchmark for Robot 'Learning by Doing'
机器人 “会做” 之后，下一步是 “会学”：清雁科技联合多家机构发布 ReAct-Manip 评测集
On October 31, 2026, at the main forum of the 2026 AI+ Conference, Qingyan Technology (Beijing) Co., Ltd. and a consortium including The Chinese University of Hong Kong (Shenzhen), Beijing Institute of Technology, Shenzhen Hetao College, AIRS, Beijing Artificial Intelligence Association, and Jiuyuan Intelligent Computing System Ecosystem Alliance released the ReAct-Manip benchmark. This industry-first evaluation set shifts focus from single-task completion to continuous learning within tasks, assessing whether robots can improve performance through interaction feedback when facing unseen physical instances.
Source: Chinese robotics — Research 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
success attempts
177
View original evidence
初步结果：GPT-6展现任务完成能力，持续适应仍有提升空间 在目前公开的测试结果中，GPT-6 Astra + primitive Harness 在 ReAct-Manip 练习阶段480次尝试中成功 177次，成功率为 36.88%；考试阶段240次尝试中成功 130次，成功率提升至54.17%，考试综合得分为 52.62。
Open source S19success attempts
130
View original evidence
初步结果：GPT-6展现任务完成能力，持续适应仍有提升空间 在目前公开的测试结果中，GPT-6 Astra + primitive Harness 在 ReAct-Manip 练习阶段480次尝试中成功 177次，成功率为 36.88%；考试阶段240次尝试中成功 130次，成功率提升至54.17%，考试综合得分为 52.62。
Open source S19total attempts
480
View original evidence
初步结果：GPT-6展现任务完成能力，持续适应仍有提升空间 在目前公开的测试结果中，GPT-6 Astra + primitive Harness 在 ReAct-Manip 练习阶段480次尝试中成功 177次，成功率为 36.88%；考试阶段240次尝试中成功 130次，成功率提升至54.17%，考试综合得分为 52.62。
Open source S19total attempts
240
View original evidence
初步结果：GPT-6展现任务完成能力，持续适应仍有提升空间 在目前公开的测试结果中，GPT-6 Astra + primitive Harness 在 ReAct-Manip 练习阶段480次尝试中成功 177次，成功率为 36.88%；考试阶段240次尝试中成功 130次，成功率提升至54.17%，考试综合得分为 52.62。
Open source S19Reported success rate
36.88%
View original evidence
初步结果：GPT-6展现任务完成能力，持续适应仍有提升空间 在目前公开的测试结果中，GPT-6 Astra + primitive Harness 在 ReAct-Manip 练习阶段480次尝试中成功 177次，成功率为 36.88%；考试阶段240次尝试中成功 130次，成功率提升至54.17%，考试综合得分为 52.62。
Open source S19Reported success rate
54.17%
View original evidence
初步结果：GPT-6展现任务完成能力，持续适应仍有提升空间 在目前公开的测试结果中，GPT-6 Astra + primitive Harness 在 ReAct-Manip 练习阶段480次尝试中成功 177次，成功率为 36.88%；考试阶段240次尝试中成功 130次，成功率提升至54.17%，考试综合得分为 52.62。
Open source S19composite score
52.62
View original evidence
初步结果：GPT-6展现任务完成能力，持续适应仍有提升空间 在目前公开的测试结果中，GPT-6 Astra + primitive Harness 在 ReAct-Manip 练习阶段480次尝试中成功 177次，成功率为 36.88%；考试阶段240次尝试中成功 130次，成功率提升至54.17%，考试综合得分为 52.62。
Open source S19rounds per task
15
- paper: not_reported
What remains unknown
Not established in the collected evidence: Environment, Control, Data origin.
Dataset licensing and commercial-use rights are not established here.
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.
初步结果：GPT-6展现任务完成能力，持续适应仍有提升空间 在目前公开的测试结果中，GPT-6 Astra + primitive Harness 在 ReAct-Manip 练习阶段480次尝试中成功 177次，成功率为 36.88%；考试阶段240次尝试中成功 130次，成功率提升至54.17%，考试综合得分为 52.62。
Open source S19
每个用例包含15个回合测试，在15个回合中，前 10 次为练习，后 5 次为考试。
Open source S14
Implications for data suppliers
RoboSignal interpretation and collection questions, not statements of buyer demand.
- Confirm the required data type and collection setting with the buyer; this source does not establish a complete collection specification.
- Compare the reported units and scope before using these quantities in a budget. Recording hours, sensor-hours and trajectories are different measures.
- Validate demand and acceptance criteria with a buyer before scaling. Publication, popularity and a research result do not establish a purchase commitment.
Source：Chinese robotics — Research news discovery · fashion.ce.cn