USTC Team Puts a 'Reverse Thinking' World Model into Robots, Secures Millions in Funding in Just Three Months
中科大团队把会「倒推」的世界模型塞进了机器人，成立三个月拿下数千万融资！
Bai Ze Tong Jing is a less than six-month-old company based in Hefei, and its developed Causal World Model AWM enables robots to perform causal reasoning about environmental changes. AWM significantly improves data efficiency and task success rates by learning object, motion, and interaction patterns from unlabeled videos.
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
≈92%
Reported accuracy
68%
memory usage
2.56
- paper: 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.
在 Push-T 机器人操作任务上，达到 75% 的成功率， AWM 只用了 2,000 条 训练轨迹；对比方法 Causal-JEPA 需要 14,728 条 ，是前者的 7 倍多。
Open source S13
当训练数据同样加到 14,728 条时， AWM 的成功率达到了 约 92% 。
Open source S15
论文标题 ： Abductive World Modeling via Causal Representation Learning 论文地址： https://arxiv.org/abs/2609.36985 代码地址： https://github.com/baiz-tech/AWM 实验显示： 与 V-JEPA 2 相比， AWM 在物理结果预测、事件推理、动作理解三类任务上均有提升，其中动作理解 Top-1 准确率提升 68% ； 遮挡画面中的关键物体时，模型判断的变化是遮挡无关物体时的 3.9 倍。
Open source S18
团队在 AMD Radeon 边缘集成显卡和 SSV2 数据集上的测试显示： 内存占用降低 2.56 倍，推理频率提升 94.6% ，精度损失不到 1% 也就是说， 同一块边缘芯片上，模型占的地方少了一大半，跑得快了将近一倍，判断的准确度却几乎没变。
Open source S24
Source：Machine Heart — Robotics on WeChat · mp.weixin.qq.com