SplAgger: Split Aggregation for Meta-Reinforcement Learning

Task inference sequence models are beneficial in meta-RL, even without task inference objectives.