Hierarchical RL sample projects:
- (2-3 ppl) Implement paper Reinforcement Learning with Competitive Ensembles of Information-Constrained Primitives.
- (1-2 ppl) Add Hindsight experience replay to Data-Efficient Hierarchical Reinforcement Learning (HIRO) implementation. Compare with simple HIRO.
- (2-3 ppl) Compare with Hierarchical Actor-Critic (HAC)
- (1-2 ppl) Add intrinsic motivation to FuN. How it affects performance on sparse reward problems like Montezuma's Revenge?
- (2-3 ppl) Based on this paper, add multi-step reward learning and temporally-extended exploration to a non-HRL agent. Compare its performance with HRL methods (e.g. FuN on Atari, HIRO on continuous actions envs or even MLSH against transferable learning properties).
- (1 ppl) Reimplement paper [Meta Learning Shared Hierarchies](https://arxiv.org/abs/1710.0976