
We propose a novel model-based reinforcement learning algorithm, Dynamics Learning and predictive control with Parameterized Actions (DLPA), that achieves superior sample efficiency and asymptotic performance compared to state-of-the-art PAMDP methods.


coremsg

model-based-reinforcement-learning-for-parameterized-action-spaces-achieving-superior-sample-efficiency-and-asymptotic-performance


Model-based Reinforcement Learning for Parameterized Action Spaces: Achieving Superior Sample Efficiency and Asymptotic Performance


title_rewrite


Several major reinforcement learning algorithms can be represented within the framework of categorical cybernetics, which models them as parametrised bidirectional processes.


reinforcement-learning-algorithms-formalized-through-categorical-cybernetics


Reinforcement Learning Algorithms Formalized through Categorical Cybernetics
