[astro-ph.EP] The Ariel mission will conduct a population level survey of hundreds of exoplanet atmospheres, requiring observations to be scheduled across a large and diverse catalogue of potential targets.
![Set Transformer actor–critic architecture used by the RL policy. The input candidate set 𝑋 ∈ R 𝑁×𝐹 is first projected into a latent 𝑑-dimensional representation and processed by 𝐿 Induced Set Attention Blocks (ISABs). Learned inducing tokens provide the queries in the first attention operation of each ISAB, after which information is redistributed to the candidate tokens. The actor combines each contextualised target representation with the global mission state to produce one logit per available observation. Each logit is then converted into a probability to select a corresponding observation. The critic uses Pooling by Multihead Attention (PMA) to obtain a permutation-invariant summary of the candidate set, which is combined with the same global state to estimate 𝑉 (𝑠). -- [astro-ph.EP]](https://i0.wp.com/astrobiology.com/wp-content/uploads/2026/10/aRieL-Reinforcement.png?resize=1200%2C715&ssl=1)
[astro-ph.EP] The Ariel mission will conduct a population level survey of hundreds of exoplanet atmospheres, requiring observations to be scheduled across a large and diverse catalogue of potential targets.