MushroomRL/mushroom-rl

Environments:
    * implement Multirobot PyBullet Interface to solve issues with joint name clash

Algorithms:
    * Conservative Q-Learning
    * Policy Search:
        - Natural gradient
        - NES
        - PAPI
        
Policy:
    * Add Boltzmann from logits for traditional policy gradient methods
    
Approximator:
    * support for LSTM
    * Generalize LazyFrame to LazyState
    * add neural network generator

For Mushroom 2.0:
    * Simplify Regressor interface: drop GenericRegressor, remove facade pattern
    * vectorize basis functions and simplify interface, simplify facade pattern
    * remove custom save for plotting, use Serializable
    * support multi-objective RL
    * support model-based RL
    * Improve replay memory, allowing to store arbitrary information into replay buffer