In urdf_description/src/links/endcap.sdf.j2 the contact parameters are currently:

<contact>
      <lateral_friction value="1.0" /><!-- TODO -->
      <rolling_friction value="1.0" /><!-- TODO -->
      <spinning_friction value="0.3" /><!-- TODO -->
      <stiffness value="3000" /><!-- TODO -->
      <damping value="100" /><!-- TODO -->
</contact>

This may be changed by measuring these values.

The min voltage seems too high for the moment.

I want to find out source file about reinforcement learning algorithm and try to learn and change something, but I could not find where they are. Would you mind give me some help? Thank you agian!

To make observations more realistic, we may

• Remove the orientation from OneLeg environments,
• Figure out a way to specify a goal position for OneLeg environments without the extremity absolute position,

Python gym_kraby is not able to load URDF and STLs. We need to find a way to include these resources in setup.py.

It seems that the method to receive observation does not properly work. Need more testing.

gym_kraby:HexapodRealEnv-v0 should command the full real robot and gym_kraby:OneLegRealEnv-v0 should command only one leg of the robot.

To make these environments work, there is a need to

• Implement and test velocity command on Herkulex servomotors,
• Set all configuration parameters on Herkulex (such as torque limit),
• Check servomotor calibration values,
• Implement Magwick IMU fusion algorithm to get the orientation of the full robot.

hexapod_bullet_env.py needs a documentation page on readthedocs.

After adding a torque limit, these motors can't move anymore.

See video: https://perso.crans.org/erdnaxe/videos/hexapod_pybullet_sliding.mkv

The rubber endcap have stiffness and damping so the robot shouldn't slide. If the robot is reversed, then it no longer slides.

Having a graph or a list with all PPO implementations tested would be nice.

In urdf_description/src/joints/joint_servomotor.sdf.j2, there is currently:

    <dynamics damping="0" friction="0" /> <!-- TODO -->
    <!-- Herkulex : 361.4°/s i.e. 6.308rad/s, 12kg.cm i.e. 1.18Nm -->
    <limit effort="1.18" velocity="6.308" lower="-1.57" upper="1.57" /> <!-- TODO -->

We may need to add a damping and friction to this joint.
We also need to check that the converted effort and velocity is right.

In the hexapod_bullet_env.py , function" def reset(self) "
What do these parameters mean?
Why there are 7 rather 6 ?
what does the last parameter mean if the first 6 are positions and orientation?
`

Reset body position/orientation

p.resetBasePositionAndOrientation(
self.robot_id,
[0, 0, 0.2],
[0, 0, 0, 1],
)
`

See https://github.com/bulletphysics/bullet3/blob/master/examples/pybullet/gym/pybullet_envs/bullet/motor.py
This might require to take apart a Herkulex DRS-0101.

Hi!I don't know why I always failed down loading gym-kraby when I use "pip install gym-kraby".I'd appreciate if you can help me!Thank you!

For the moment, the simulation timestep is the same as the timestep of the environment. This may diverge the simulation is the timestep is too big.

Traceback (most recent call last):
  File "herkulex_socket.py", line 140, in <module>
    print(h.get_observations())
  File "herkulex_socket.py", line 112, in get_observations
    pos = ret[9] + ((ret[10] & 0x3) << 8)
IndexError: index out of range

This may be fixed by adding a loop on reception.