[Python] In Gymnasium, how can one run a vector environment's function in parallel similar to...

I have a custom Gymnasium environment, RLToy-v0 from the library MDP Playground. It separates out the transition function and reward function from the step function and calls them individually inside step().

I know that using Gymnasium's AsyncVectorEnv, and initialising a set of vectorised environments as envs, we can call envs.step() and envs.observation_space.sample() to run the respective individual functions of the environments in parallel. However, I want to run only the reward function, named reward_function(), of the vectorised environments in parallel. How would one go about doing that for the separated out reward function of my custom environment? Is there an easy way to just call envs.reward_function()?

Continue reading...