1 minute read

During these two weeks, I focused on learning how to use the neural network.

Since the network was not created by me, the first thing I needed to do was determine what its inputs were.

Using the following code:

for i in session.get_inputs():
    print(i.name, i.shape, i.type)

for o in session.get_outputs():
    print(o.name, o.shape, o.type)

I obtained the following result:

obs [1, 99] tensor(float)
actions [1, 29] tensor(float)

Since the robot has 29 joints, the output clearly corresponds to the commands sent to those joints. The problem was understanding the inputs.

After checking how the network was originally designed, I concluded that the inputs were:

  • IMU linear velocity
  • IMU angular velocity
  • Projected gravity
  • Joint positions
  • Joint velocities
  • Previous output
  • Desired command

The issue appeared when checking the dimensions.
If all the data obtained from MuJoCo were placed in a single array, the resulting dimension was [1, 113].

While searching online, I found that the positions and velocities returned by MuJoCo include some values that are not related to the joints themselves. After removing those values, the size became [1, 100], which is still one element larger than expected.

Since I could not find where the extra value was coming from, I tried removing the last element of the twist command. This value represents the desired facing direction, so I thought it might not be essential.

It clearly did not work.

G2_failure


Summary

  • Identified the network inputs and outputs
  • Investigated how to obtain the required data
  • Attempted to run the network

Categories:

Updated: