Humanoid Robot Catches Thrown Boxes Using Pure Vision in Three-Stage Reflex Framework
人形机器人学会接住飞来的箱子：一场视觉与反应的三阶段实验
Researchers from the University of Washington, USC, and other institutions developed Reflex, a three-stage training framework enabling a Unitree G1 humanoid robot to catch unmarked cardboard boxes using only onboard RGB-D cameras. In real-world tests, the robot achieved 65% success at 2.5 meters and 40% at 5 meters, compared to human rates of 85% and 68.3%. The approach addresses whole-body coordination, latency prediction, and visual complexity without external motion capture or reflective markers.
Source: Chinese robotics — Hardware and sensing · 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
85%
Reported trials
View original evidence
结果是,在2.5米距离下机器人接住了65%的箱子,5米距离下是40%,作为对比,三个真人在同样距离下的成功率是85%和68.3%。
Open source S7Reported success rate
68.3%
Reported trials
View original evidence
结果是,在2.5米距离下机器人接住了65%的箱子,5米距离下是40%,作为对比,三个真人在同样距离下的成功率是85%和68.3%。
Open source S7flight duration
1 seconds
View original evidence
论文里提到一个关键数字:在真实机器人实验中,箱子从扔出到到达机器人身边大约只有1秒,而摄像头采集图像到机器人做出控制指令这中间的延迟,中位数就有69毫秒。
Open source S12control latency
69
View original evidence
论文里提到一个关键数字:在真实机器人实验中,箱子从扔出到到达机器人身边大约只有1秒,而摄像头采集图像到机器人做出控制指令这中间的延迟,中位数就有69毫秒。
Open source S12
- paper: not_reported
- code: not_reported
- weights: not_reported
- dataset: 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.
结果是,在2.5米距离下机器人接住了65%的箱子,5米距离下是40%,作为对比,三个真人在同样距离下的成功率是85%和68.3%。
Open source S7
机器人达到了人类表现的76%和59%。
Open source S8
论文里提到一个关键数字:在真实机器人实验中,箱子从扔出到到达机器人身边大约只有1秒,而摄像头采集图像到机器人做出控制指令这中间的延迟,中位数就有69毫秒。
Open source S12
Source：Chinese robotics — Hardware and sensing · msn.cn