Darwin-OP simulation online adaptive reinforcement learning in VM's
MSc final project required running many experiments for Darwin-OP using Webots.
Each experiment uses a stochastic (random) approach to populate a finite search-space. Then, using Reinforcement Learning the robot tries to optimise its policy on how to stand up.
Eventually, the robot is "shown" how to stand up, and it learns even faster by imitating sequences.
You can find my Reinforcement Learning C++14 library here: http://alexge233.github.io/relearn/
This video shows many VM's running in parallel, in order to increase the number of experiments and decrease the overall time.
The eventual solution to the near-infinite search space of sensorimotor functions, was to provide imitative didactic paradigm and then require the r.learning algorithm to repeat the correct procedure/episode. Stochastic approaches would have taken a very long time.