Index

Reference JAX RL agents you can train on navix environments: PPO, PQN (a replay/target-free deep Q-network) and Dreamer (DreamerV3). Each pairs with an HParams struct and plugs into navix.experiment.Experiment. models holds the shared network pieces, including the carry-based Encoder family (swap TransformerEncoder in for a history-conditioned policy on partially observable tasks).