pretrain.tex

\begin{algorithm}[t!]
\caption{Pretrain and Fine Tune}
\begin{algorithmic}[1]
\label{alg:pretrain}
\REQUIRE Demonstrations data $\mathcal{D}$
\STATE Collect demonstration data $\mathcal{D} = \{(s_i,a_i)\}$
\STATE Initialize $\pi_\theta$ as $\max_\theta\sum_i\log\pi_\theta(a_i|s_i)$
\WHILE{not done}
\STATE Run $\pi_\theta$ to collect experience
\STATE Improve $\pi_\theta$ with any RL algorithm
\ENDWHILE
\end{algorithmic}
\end{algorithm}