thank you very much!

The PPO backend rl_games in this repo seems to be a "bad" choice for legged learning, and I am using rsl_rl intensively for my current project. If there is a need or I have some free time, this could be a planned work.

Thank you for your great work! I am very interseted in it.
I want to add unitree Go2 to RobotType, then I check the <rl-mpc-locomotion/MPC_Controller/common/Quadruped.py> where defines some physical properties of the robots, like A1,Go1. I would like to know how some parameters are determined so that I can adjust them according to Go2. Take Go1 as an example

if robotype is RobotType.GO1:
        self._abadLinkLength = 0.08
        self._hipLinkLength = 0.213
        self._kneeLinkLength = 0.213
        self._kneeLinkY_offset = 0.0
        self._abadLocation = np.array([0.1881, 0.04675, 0], dtype=DTYPE).reshape((3,1))
        self._bodyName = "trunk"
        self._bodyMass = 5.204 * 2
        self._bodyInertia = np.array([0.0168128557, 0, 0, 
                                      0, 0.063009565, 0, 
                                      0, 0, 0.0716547275]) * 5
        self._bodyHeight = 0.26
        self._friction_coeffs = np.ones(4, dtype=DTYPE) * 0.4
        # (roll_pitch_yaw, position, angular_velocity, velocity, gravity_place_holder)
        self._mpc_weights = np.array([1.0, 1.5, 0.0,
                                     0.0, 0.0, 50,
                                     0.0, 0.0, 0.1,
                                     1.0, 1.0, 0.1,
                                     0.0], dtype=DTYPE) * 10

What do the following parameters mean?
self._abadLinkLength
self._abadLocation
How are this related to the URDF file in assets folder, or can it be determined from the URDF?
why does the self._bodyMass Multiply by 2?
Does the self._mpc_weights is just an initial value?

I do have a parallelized, torch-based version implemented but with minor bugs when solving the QP in qpth. If there is a need or I have some free time, this could be a planned work.

Although I know Yuxiang has already had one parallelized QP controller well implemented lol.

Hello! It is really a great work! When I run MPC controller, I find that the robot's step frequency is very high. I want to know how can I change it, thank you!

Hi @silvery107,

Thanks a lot for releasing the nice project. During studying the code base, I met some issues and want to discuss them with you.

1. In the update function within /MPC_Controller/common/StateEstimator.py, the current code for updating vBody and omegaBody is as follows:
   self.result.vBody = self.result.rBody @ self.result.vWorld
   self.result.omegaBody = self.result.rBody @ self.result.omegaWorld
   However, I believe the correct transformations should be:
   self.result.vBody = self.result.rBody.T @ self.result.vWorld
   self.result.omegaBody = self.result.rBody.T @ self.result.omegaWorld
   This is based on the assumption that we want to transform the linear and angular velocities from the world frame to the body frame, where v_b = R_b_to_s @ v_s?

2. Additionally, I noticed a minor drift in the yaw-axis and the y-axis when the command is set to zero, as shown in the following video link:
   https://github.com/silvery107/rl-mpc-locomotion/assets/71823391/0c03f991-0c5c-46b8-b499-c07a644657c9
   Could you please give some suggestions on how to improve it?

Thanks a lot for your time and look forward to your response~~~

I am trying to do sim2real on Go1 based on your code.
In your code, controller_dt = 0.01 # in sec, but I am curious whether controller_dt was set to 0.01 in the actual robot.

Hello, I would like to ask if you can participate in this project if you can successfully install Isaac gym but cannot run Isaac sim correctly. Isaac Sim prompts that the graphics card cannot be found, but I made sure to install the relevant driver myself.
In addition, I would like to ask if there is terrain information such as scanning point data in the framework that participates in the training. If so, where should I start to learn the status configuration of the environment information?

Hi , bro . Recently I am trying to train a PPO policy in Isaacgym and deploy to AliengoRobot , but I see your stiffness=0 damping =0 , how can you control your robot to move ?

I have tested this codebase under python 3.8 earlier, and didn't see much incompatability. The only limit of python version seems to be the bindings from Isaac Gym. I will boost the python version after a more thorough test.

Thank you very much for your sharing, which gives me a lot of inspiration.
I want to deploy this controller on the robot I built later. Could you recommend a real robot deployment scheme for reference?
Thanks a lot.

Hi, I have successfully trained a PPO model in isaacgym and deploy to ros and gazebo . My model walks well in gazebo , but I don't know how to deploy to a real A1 robot . I read the unitree's official demo , their demo is c++ , I don't know how to deploy in c++ , do you have python to control the robot by 12 joints like in gazebo ?

Hi @silvery107,
Thanks for this very rich work.

Could you list some guidelines for deploying your code to Unitree Go1?

Thanks for the feedback.

Lucas

I somehow missed the correct version of osqp when I first published this repo. The osqp folder sould be replaced as a submodule @ its corresponding commit.

thank you!!!!