Felix Chalumeau

I am not sure the DQN example (model-zoo/contrib/games/differentiable-programming/cartpole/DQN.jl) really fits with the original DQN paper Playing Atari with Deep Reinforcement learning (https://arxiv.org/abs/1312.5602) as the original algorithm perform the experience replay...

Ant with a trap of boxes (similar to issue #136)

9

Hi! I am trying to extend the [Ant](https://github.com/google/brax/blob/main/brax/envs/ant.py) basic brax locomotion environment by simply adding a Trap in front of the Ant. To do so, I took the _SYSTEM_CONFIG of...

Hi all, At the moment, PointMaze's action space is [-0.1, 0.1]² whereas all the other environments have an action space of [-1, 1]**n Wouldn't it be better to have the...

We observed strange results on several environments. It seems that the behavior descriptors are not properly evaluated in some cases. More information in the related PR.

Tackle the "stochastic" issue while training: taking the difficulty of an episode into account.

2

The title might not be explicit enough, let me explain this point. When training an agent to play mountainCar, CartPole, CarRacing etc... the best scores he can get are rather...

Representable variables

2

Inspired by #71 . In the same way that we can choose if a variable is branchable, it could be interesting to have the flexibility to choose if a variable...

I have got a doubt on the proper use of the opt states in the PG mutation of PGAME. I fear the opt state is not properly re-initialised from one...

Add algorithm CMA-ME

1

Related to issue #80 Done: - implement the improvment emitter - implement the random emitter - implement the opt emitter - implement the pool of emitters - example notebook -...

Adding tasks and default scoring functions

1

Let's add tasks to QDax (rastrigin, sphere, arm, etc...) and provide default scoring functions for them as well as for Brax environments.

Link to the papers: - https://ieeexplore.ieee.org/abstract/document/9736995 - https://dl.acm.org/doi/abs/10.1145/3321707.3321804