Study Suggests GPT-6 Astra's Robotic Capabilities May Stem from Specific Training
GPT-6 Astra的机器人能力被扒：可能接受过专门训练
A recent study indicates that GPT-6 Astra's robotic capabilities are not emergent but result from specific training on robot data. Researchers tested the model on forward and inverse kinematics tasks using both standard Franka Panda parameters and synthetic Denavit-Hartenberg parameters. The model failed to adapt its predictions when given new DH parameters for a different robot structure, suggesting it memorized specific kinematic knowledge rather than understanding general physical laws. This contrasts with models that successfully generalized to unseen robot configurations.
Source: Chinese robotics — Dataset and collection 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.
- Environment
- SimulationOpen source S6
- paper: not_reported
Source excerpts and review record
Automatically extracted; no manual editorial approval recorded.
发布后，NVIDIA 将 GPT-6 Astra 引入机器人仿真与设计工作流。
Open source S6
来源：https://x.com/nvidiaomniverse/status/2118450742176698700 另一项演示借助 Unitree G1 人形机器人，Astra 可以在一个从未见过的房间里自主行动，记住不同物体的位置，跨越整个房间完成整理任务。
Open source S8
在一项演示中，开发者使用 Astra、PTC Onshape 和 NVIDIA Isaac Sim，建立汽车悬架的 3D 模型，调整扳手的形状与位置，使其能够触及悬架上的螺栓，最终在仿真环境中演示机器人使用扳手拆卸部件的过程。
Open source S7
来源：https://x.com/SwannAiden/status/2118741112227774702 该研究指出， GPT-6 Astra 的机器人能力并非涌现出来的，OpenAI 在训练过程中使用了专门的机器人数据。
Open source S16
但研究者发现，GPT-6 Astra 等三个模型即使拿到了另一台七自由度机器人的 DH 参数，给出的预测仍然与 Franka Panda 机械臂的运动学特征相符。
Open source S23
一类使用 Franka Panda 机械臂的运动学参数，另一类则使用合成生成的机器人运动学参数。
Open source S17
Source：Chinese robotics — Dataset and collection discovery · news.qq.com