Sven Mika b10d5533be [RLlib] Issue 20920 (partial solution): contrib/MADDPG + pettingzoo coop-pong-v4 not working. (#21452) | 2 years ago | |
---|---|---|
.. | ||
tests | ed85f59194 [RLlib] Unify all RLlib Trainer.train() -> results[info][learner][policy ID][learner_stats] and add structure tests. (#18879) | 3 years ago |
README.md | eae7a1f433 [RLLib] Readme.md Documentation for Almost All Algorithms in rllib/agents (#13035) | 3 years ago |
__init__.py | 4d7bd8c892 [RLlib] Implementation of "Model-based Meta Policy Optimization" (MB MPO) (#9409) | 4 years ago |
mbmpo.py | b10d5533be [RLlib] Issue 20920 (partial solution): contrib/MADDPG + pettingzoo coop-pong-v4 not working. (#21452) | 2 years ago |
mbmpo_torch_policy.py | f82880eda1 Revert "Revert [RLlib] POC: Deprecate `build_policy` (policy template) for torch only; PPOTorchPolicy (#20061) (#20399)" (#20417) | 2 years ago |
model_ensemble.py | cf21c634a3 [RLlib] Fix deprecated warning for torch_ops.py (soft-replaced by torch_utils.py). (#19982) | 3 years ago |
utils.py | ce96b03b07 [RLlib] MB-MPO cleanup (comments, docstrings, type annotations). (#11033) | 4 years ago |
Code in this package is adapted from https://github.com/jonasrothfuss/model_ensemble_meta_learning.
MBMPO is an on-policy model-based algorithm. On a high level, MBMPO is model-based MAML. On top of MAML, MBMPO learns an ensemble of dynamics models. MBMPO trains the dynamics models with real-life data and the actor/critic networks with fake data generated by the dynamics models. The actor and critic are updated via the MAML algorithm. For the distributed execution plan, MBMPO alternates between training the dynanmics model and training the actor and critic network.
More details can be found here.
MBMPO.