Jack Anders Smitterberg and Tan Chen have released an open-source graphical tool that records proprioceptive data from Unitree's Go1 quadruped, according to a paper on arXiv. It logs joint states, inertial measurements, foot force and gyroscope readings without the Go1's native ROS1 software, which is no longer supported.
The authors checked 51 recordings made with the tool for sampling rate and repeated samples. As a deliberately modest test, a terrain classifier trained on the data reached about 38% accuracy across ten terrain types, against 10% by chance. They call the result weak and say it only shows that the signals carry information about the terrain.



