Reinforcement learning environments with musculoskeletal models
https://www.crowdai.org/challenges/nips-2017-learning-to-run
CANDIDATE ALGORITHMS:
Depp Deterministic Policy Gradient---DDPG(https://arxiv.org/abs/1509.02971)
Recurrent Deterministic Policy Gradient---RDPG(https://arxiv.org/pdf/1512.04455.pdf)
Trust
MOTIVATION:
In this case, we are manipulating the muscles rather than the velocity of the body parts, i.e., if I understood the problem correctly, we are changing the acceleration, which should be considered a second-order markov chain. LSTM and recurrent network might be able to capture the long short term dependencies.
FOlDERS AND FILES:
DDPG: The standard implementation of ddpg in tensorflow, following the pesudocode in the paper.
RDPG: The recurrent version of the ddpg in tensorflow(There are complications that have not been resolved)
DEPENDENCIES: Tensorflow numpy math