Required prerequisites
What version of OmniSafe are you using?
0.5.0
Problem description
The docstring for PIDLagrangian.pid_update states the update rule as:
$\lambda_{t+1} = \lambda_t + (K_p e_p + K_i \int e_p dt + K_d \frac{d e_p}{d t}) \eta$
However, this does not match the actual implementation: there is no learning rate $\eta$ and the update is not incremental ($\lambda_{t+1} = \lambda_t + \dots$).
I suggest to replace the docstring formula with the following:
$\lambda_{t+1} = \max(0,K_p e_p + K_i \int e_p dt + K_d \frac{d e_p}{d t})$
Required prerequisites
What version of OmniSafe are you using?
0.5.0
Problem description
The docstring for
PIDLagrangian.pid_updatestates the update rule as:However, this does not match the actual implementation: there is no learning rate$\eta$ and the update is not incremental ($\lambda_{t+1} = \lambda_t + \dots$ ).
I suggest to replace the docstring formula with the following:
$\lambda_{t+1} = \max(0,K_p e_p + K_i \int e_p dt + K_d \frac{d e_p}{d t})$