Zhejiang University and TeleAI Propose GR-BC for Precise Robotic Manipulation
CoRL 2026｜浙大联合TeleAI提出GR-BC，用图搜索从机器人试错中找出有效动作
A team from Zhejiang University, in collaboration with China Telecom AI Research Institute (TeleAI) and Gamma Intelligence, proposed Graph-Reweighted Behavior Cloning (GR-BC). The method uses graph-based trajectory analysis to identify effective action segments from imperfect demonstrations. Evaluated on four fine-grained tasks including DDR4 memory installation, GR-BC achieved an average success rate of approximately 91% using the X-VLA backbone.
Source: Machine Heart — Robotics on WeChat · 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
≈91%
Reported trials
View original evidence
图 5｜GR-BC 的离线迭代表现 实验结果：四项精细操作，平均成功率约 91% 真机实验覆盖四孔零件装配、显卡插装、双螺钉拧紧和内存条插装。
Open source S17Reported success rate
98%
Reported trials
View original evidence
GR-BC 在上述四项任务上的最终成功率依次为 98%、93%、89% 和 83%， 平均约 91%；类 DAgger 等权聚合基线平均为 64%，相差约 27 个百分点。
Open source S19Reported success rate
93%
Reported trials
View original evidence
GR-BC 在上述四项任务上的最终成功率依次为 98%、93%、89% 和 83%， 平均约 91%；类 DAgger 等权聚合基线平均为 64%，相差约 27 个百分点。
Open source S19Reported success rate
89%
Reported trials
View original evidence
GR-BC 在上述四项任务上的最终成功率依次为 98%、93%、89% 和 83%， 平均约 91%；类 DAgger 等权聚合基线平均为 64%，相差约 27 个百分点。
Open source S19Reported success rate
83%
Reported trials
View original evidence
GR-BC 在上述四项任务上的最终成功率依次为 98%、93%、89% 和 83%， 平均约 91%；类 DAgger 等权聚合基线平均为 64%，相差约 27 个百分点。
Open source S19evaluation trials per task
30
Reported trials
- paper: not_reported
- code: Reported pending Open source S9
- weights: not_reported
- dataset: not_reported
- code: 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.
图 5｜GR-BC 的离线迭代表现 实验结果：四项精细操作，平均成功率约 91% 真机实验覆盖四孔零件装配、显卡插装、双螺钉拧紧和内存条插装。
Open source S17
GR-BC 在上述四项任务上的最终成功率依次为 98%、93%、89% 和 83%， 平均约 91%；类 DAgger 等权聚合基线平均为 64%，相差约 27 个百分点。
Open source S19
每项任务固定评测 30 条成功、30 条失败轨迹。
Open source S22
论文标题：Graph-Reweighted Behavior Cloning with Human Interventions for Precise Robotic Manipulation 论文链接： https://arxiv.org/pdf/2610.20504 项目主页： https://gr-bc.github.io 代码链接： https://github.com/Gamma-Robot/GR-BC 核心方法：用图检索进度，再为动作加权 图 1｜GR-BC：构图、反向搜索、加权训练与新轨迹回流。
Open source S9
Source：Machine Heart — Robotics on WeChat · mp.weixin.qq.com