Algorithm 2 VIB based meta-reinforcement learning testing algorithm.

1:Input:${\left\{{\mathcal{T}}_n\right\}}_{m=1,\dots ,N}\sim p(\mathcal{T})$: Meta-testing task set; $\theta$: meta-training policy network; $\omega$: meta-training latent space encoder network. 2:Initializing the trajectory: $e^{\mathcal{T}}=\left\{\right\}$ 3:for $k=1,\cdots ,N$ do 4:      Latent space inference: $z\sim E_{\omega }\left(\left.z\right|e^{\mathcal{T}}\right)$ 5:      Using current policy ${\pi }_{\theta }\left(\left.a\right|s,z\right)$ to interact with each task and obtain ${\mathcal{D}}^k$ 6:      Sample accumulation: $e^{\mathcal{T}}=e^{\mathcal{T}}\cup {\mathcal{D}}^k$ 7:      Evaluating empirical discounted return for each task. 8: end for