Learning to Walk Using Reinforcement Learning
We are tasked with training a bipedal walker bot to walk
We use the Box2D environment provided by openAIGym
Input parameters of the environment per step: Four real numbers between -1 and 1
Output values per step:
- Observation, an array of 24 numbers representing the current state
- Reward value
- A Boolean representing whether or not the bot has fallen
- Debugging information (not used in this experiment)
We implement active reinforcement learning algorithms in order to teach the bot how to move effectively without falling.
