You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Currently, G1's controllers are only validated in simulation, specifically in Mujoco. tbai uses a fork of Unitree's mujoco simulator, thus the simulation interface and the real robot interface should not differ as much. In fact, the only thing one should need to change in order to deploy all tbai's implemented controllers on G1 is the networkInterface and the unitreeChannel attributes in the G1RobotInterfaceArgsstruct, which serves as the only argument to G1RobotInteface's constructor.
Go2's controllers already have been validated, deployment launch files can be found in tbai's ROS repository, here. We would like to create a similar deployment ros package for g1 deployment with launch files for all currently implemented controllers, including the static controller, the walking controller, the dance controller and finally the beyond mimic controller.
With the launch files and other helper scripts in place, we would like to create a set of videos showcasing the behavior of each and every single one of the controllers mentioned above. The movements should look a little like this:
Stand Controller
Walking Controller
walk.mp4
Dance Controller
dance2.mp4
Beyond Mimic Controller
beyond.mp4
You can run the above controller by installing tbai_ros, it's a matter of running no more than three commands. After that, one should be able to simulate g1 by running just g1_mujoco. Proceed by picking your desired controller in the virtual joystick's GUI. The robot will initially be spawned in the air and suspended from a virtual elastic band. You can loosen this elastic band by pressing 8 on your keyboard, deactivating it completely is as simple as pressing 9.
When doing experiments, one should be extremely careful and validate only one thing at a time. It may as well happen that the joint indeces do not match indeces on the real robot, therefore visualizing G1 in either rerun or rviz will likely prove extremely helpful too. In case of questions or comments, I'll be happy to provide any support, I am available 24/7.
Currently, G1's controllers are only validated in simulation, specifically in Mujoco. tbai uses a fork of Unitree's mujoco simulator, thus the simulation interface and the real robot interface should not differ as much. In fact, the only thing one should need to change in order to deploy all tbai's implemented controllers on G1 is the
networkInterfaceand theunitreeChannelattributes in theG1RobotInterfaceArgsstruct, which serves as the only argument toG1RobotInteface's constructor.Go2's controllers already have been validated, deployment launch files can be found in tbai's ROS repository, here. We would like to create a similar deployment ros package for g1 deployment with launch files for all currently implemented controllers, including the static controller, the walking controller, the dance controller and finally the beyond mimic controller.
With the launch files and other helper scripts in place, we would like to create a set of videos showcasing the behavior of each and every single one of the controllers mentioned above. The movements should look a little like this:
Stand Controller
Walking Controller
walk.mp4
Dance Controller
dance2.mp4
Beyond Mimic Controller
beyond.mp4
You can run the above controller by installing
tbai_ros, it's a matter of running no more than three commands. After that, one should be able to simulateg1by runningjust g1_mujoco. Proceed by picking your desired controller in the virtual joystick's GUI. The robot will initially be spawned in the air and suspended from a virtual elastic band. You can loosen this elastic band by pressing8on your keyboard, deactivating it completely is as simple as pressing9.When doing experiments, one should be extremely careful and validate only one thing at a time. It may as well happen that the joint indeces do not match indeces on the real robot, therefore visualizing G1 in either rerun or rviz will likely prove extremely helpful too. In case of questions or comments, I'll be happy to provide any support, I am available 24/7.