roboterax / humanoid-gym Goto Github PK

Hi just realized that in the init_buffers() function the dimension of self.rigid_states is hard-coded, this might cause problems if other robots are used and has different number of bodies

https://github.com/roboterax/humanoid-gym/blob/main/humanoid/envs/base/legged_robot.py#L455

The Isaac Gym Preview 4 version used in the project is older and has many known problems, such as contact force, coordinate system, model solution, etc.

I noticed that legged_gym has been ported to the new simulation platform orbit in January this year. Will the author consider porting in the future?

Could you do a brief performance comparison, like GPU usage? Appreciate it.

how do i do if i don't have the xbot real robot.
and when did you want the robot to do some manipulation task?

Hello author! Your open source code can very well solve the walking problem of our humanoid robot. But after several days of trying, we have a problem that we haven't solved yet.
Our model can walk normally when training "plane", but it cannot walk normally when training "trimesh", and even forgets how to walk. We increased the number of iterations to 10,000 before the robot could barely walk. I would like to ask which part of the modification have we overlooked? Or what should I pay attention to when changing models?

I see that the code has many domain randomization Settings, as shown in the figure, but only a small part of them are used. Can such Settings ensure the robustness of sim2real

Hey, thanks for the great contribution!
I have question regarding urdf to mjcf mapping. How did you transfer the robot model between these two formats?

The function that compute reference state is very hard to understand. Why ref_dof_pos is related to sin of phase? What does the scale1/2 stand for? Why only 2,3,4 and 8,9,10 are specified?

I really appricate your help and answer.

    def compute_ref_state(self):
        phase = self._get_phase()
        sin_pos = torch.sin(2 * torch.pi * phase)
        sin_pos_l = sin_pos.clone()
        sin_pos_r = sin_pos.clone()
        self.ref_dof_pos = torch.zeros_like(self.dof_pos)
        scale_1 = self.cfg.rewards.target_joint_pos_scale
        scale_2 = 2 * scale_1
        # left foot stance phase set to default joint pos
        sin_pos_l[sin_pos_l > 0] = 0
        self.ref_dof_pos[:, 2] = sin_pos_l * scale_1
        self.ref_dof_pos[:, 3] = sin_pos_l * scale_2
        self.ref_dof_pos[:, 4] = sin_pos_l * scale_1
        # right foot stance phase set to default joint pos
        sin_pos_r[sin_pos_r < 0] = 0
        self.ref_dof_pos[:, 8] = sin_pos_r * scale_1
        self.ref_dof_pos[:, 9] = sin_pos_r * scale_2
        self.ref_dof_pos[:, 10] = sin_pos_r * scale_1
        # Double support phase
        self.ref_dof_pos[torch.abs(sin_pos) < 0.1] = 0

        self.ref_action = 2 * self.ref_dof_pos

Dear author.

Thank you for your awesome work, How does policy handle privileged observations in real-world?
and Why do the privileged observations have 73 dimensions in Table 3 but a different number in Table 1 in the paper?What does privileged observation involve?

Hello, I am really interested in this project and would like to have implement it.

During installation setup, you mentioned using python 3.8 in the first step, however, in step 3 it does not seem to be compatible with it, here is the output:

Could not solve for environment specs
The following packages are incompatible
├─ pin-1 is installable and it requires
│ └─ python 3.11.* , which can be installed;
└─ torchvision 0.14.1 is not installable because there are no viable options
├─ torchvision 0.14.1 would require
│ └─ python >=3.10,<3.11.0a0 , which conflicts with any installable versions previously reported;
├─ torchvision 0.14.1 would require
│ └─ python >=3.7,<3.8.0a0 , which conflicts with any installable versions previously reported;
├─ torchvision 0.14.1 would require
│ └─ python >=3.8,<3.9.0a0 , which conflicts with any installable versions previously reported;
└─ torchvision 0.14.1 would require
└─ python >=3.9,<3.10.0a0 , which conflicts with any installable versions previously reported.

It would be greatly appreciated if we could have a further discussion regarding the entire project, please contact me if you are interested, my email address is [email protected]
Both Mandarin and English are fine with me.

Thank you in advance

All the best
Zhengting Li

Hello!

In the config file I found that kp = 15 and kd = 10 for the ankles. The kp/kd pair seems mismatching compared to other joints. Is it a typo for kp = 150 or an intentional design?