The mutable robot object instance in the default init parameter caused the default robot object to be shared between multiple instances of environments.
For example, if you use stable-baselines3 (where they did the training and evaluation implementation RIGHT),
they have separate train and eval environment instantiated and the robot gets shared between the two.
https://github.com/bulletphysics/bullet3/issues/2352 : Fixes this problem.
The mutable robot object instance in the default init parameter caused the default robot object to be shared between multiple instances of environments. For example, if you use stable-baselines3 (where they did the training and evaluation implementation RIGHT), they have separate train and eval environment instantiated and the robot gets shared between the two.