This repo builds on the code from rl_locomotion
Completely follow the steps of the original warehouse for installation and use
Depending on each individual's environment,
there may be some minor issues when configuring the environment,
but you can solve your problems through Google
Reset rewards to correct unstable standing issues. see video:
Crop the Go1 description file provided by Unitree to adapt to the old version of raisim
Updated jt_mean_pos in this file and this
I conducted experimental comparisons, including modifying different hyperparameters and adjusting rewards.
The blue curve represents the results of training the initial code under a1,
while the other curves represent the results of training under go1.
The purple and yellow curves represent the training results when the RL_coeff parameters are 0.2 and 0.5, respectively,
with a greater error compared to the original curve.
The green curve is the result of setting the RL parameter to 0.8 and modifying the reward.
It can be seen that only the green curve is closest to the original training result
