Code not running on Untree A1 robot

[Shifters1]

Thank you for your work. I have been trying to deploy your policy on a unitree A1 robot but it doesnt work. I have followed all your instruction as specified on your page. I editted the Lcm_position.cpp to suit UDP requirement of A1 and i have compiled and updated the lcm_position file. I have verified the deployemt.tar file to ensure that the check md5sum deployment_image.tar has ca70e4a10063fa39f026d32b1fe5d88b. While troubleshooting this problem, I came across a similar issue on your webpage https://github.com/Improbable-AI/walk-these-ways/issues/1] and it still doesnt provide any solution.

Each time try to run this code, i keep getting stuck at the image below

Please help with a solution to this!!

[gmargo11]

Hi @Shifters1 ,

Unfortunately, I don't have access to an A1 robot, so I can only give some high level suggestions.

After running the lcm_position file, try moving the joints. Do they move freely or resist motion? If they move freely, there is a problem with the lcm_position file and its usage of the unitree_legged_sdk. If they resist motion, there is a problem with the communication between lcm_position and the policy code.

If the former, make sure you can run the low-level examples from the unitree_legged_sdk repo.

If the latter, try entering the docker container in interactive mode using sudo make run instead of sudo make autostart. Then run the deployment script inside docker in interactive mode: cd /home/isaac/go1_gym/ && python3 setup.py develop && cd go1_gym_deploy/scripts && ls && python3 deploy_policy.py. Do you see any additional output?

[Shifters1]

Thank you for your prompt response. I successfully resolved the issue by commenting out the code section that required me pressing R2, enabling the policy to function on the Unitree A1 robot. However, a new problem arose as the remote control is now non-responsive in low level, although it works generally. Additionally, i tested the default example_joystick on the SDK and it showed no response when pressing A. As a result, I currently lack the capability to control the gaiting and fully utilise your work.

Furthermore, I noticed that terminating the policy any time is running on the robot is not feasible, it always requires shutting down the robot.

Could you help with possible solutions?

Many thanks Gabe

[gmargo11]

Hmm, if example_joystick is not working, this is a problem with Unitree's SDK on A1. Unfortunately, I can only suggest opening an issue in https://github.com/unitreerobotics/unitree_legged_sdk or emailing Unitree support.

By the way, are you using v3.3.1 of the unitree_legged_sdk? I noticed in https://github.com/unitreerobotics/unitree_legged_sdk#readme that this version is required for A1.

-Gabe

[Shifters1]

Thank you Gabe for your quick response and suggested solutions. We are currently using v3.3.1 and have discovered that the problem is related to the SDK. We have already contacted the Unitree support service.

However, there is one more question that you omitted. We're having trouble stopping the policy once it's starts running on the robot which always necessitate total shutdown of the robot. Is there a known solution to this issue?

[gmargo11]

1. If you are running docker in interactive mode as described above, you can kill the controller with Ctrl-C in the docker window.
2. If you just want to stop the controller and transition the robot into a passive standing pose, the button is [R2]. But since your controller's not working, this won't work for you yet.
3. By the way, because this is a common request, I will be pushing a new pretrain-v1 policy soon that automatically transitions to a learned standing mode when the velocity command is low. This will mean the robot doesn't trot in place anymore.

-Gabe

[keving2733]

Hello @Shifters1,

I am also trying to deploy a trained policy on an A1. What specifically needs to change in Lcm_position.cpp to suit UDP requirement of A1? Is there anything else that needs to be changed in this walk_these_ways package? I've looked at the unitree_legged_sdk for the A1 and not sure if that is supposed to help, but if you could point me to any other resources I would greatly appreciate it, thanks!

-Kevin

[Shifters1]

Hi [@keving2733],

Here's how we deployed ours: I downloaded and built SDK Version 3.3.1 on the robot. Then, I updated the sport version to 1.19 on the Upboard (the board toward the hind leg). I copied the lcm_position.cpp from "walk these ways" to the example file of the SDK. I commented out any parts related to battery or battery state. This is because battery management/bms wasn't captured in V 3.3.1 comm.h. I also changed the robot's name to A1 and left UDP settings as level in the lcm_position.cpp. Dont forget to add lcm_position to the CMakeList.txt

Once the build was successful, I copied both lcm_position.cpp and lcm_position from the build to the sdk_bin folder on "walk these ways". This should fix it.

[keving2733]

Hello @Shifters1,

You copied the lcm_position.cpp into the "example" folder correct? I am also unable to identify where in "lcm_position.cpp" the parts about the battery are. I also notice in the sdk in the "example" directory the example files seem to name the robot "Go1". I may be looking in the wrong place but if you happen to know where it is where the robot is named, I would greatly appreciate it. Thanks!

Best,

Kevin Garcia

[Shifters1]

Hi Gabe ,

I would like to know when you plan to release the pretrain-v1 policy. Thank you

3. By the way, because this is a common request, I will be pushing a new pretrain-v1 policy soon that automatically transitions to a learned standing mode when the velocity command is low. This will mean the robot doesn't trot in place anymore.