diff --git a/dopamine/agents/rainbow/rainbow_agent.py b/dopamine/agents/rainbow/rainbow_agent.py index 00b6d138..7e3e4ebb 100644 --- a/dopamine/agents/rainbow/rainbow_agent.py +++ b/dopamine/agents/rainbow/rainbow_agent.py @@ -194,7 +194,8 @@ def _build_replay_buffer(self, use_staging): stack_size=self.stack_size, use_staging=use_staging, update_horizon=self.update_horizon, - gamma=self.gamma) + gamma=self.gamma, + observation_dtype=self.observation_dtype.as_numpy_dtype) def _build_target_distribution(self): """Builds the C51 target distribution as per Bellemare et al. (2017).