clone and install latest metaworld from source (but with mujoco>=2.3:
git clone https://github.com/balloch/Metaworld.git && cd Metaworld && pip install . && cd ..
pip install continual world
only really tested with run_cl.py so far, and haven't tested the CL methods yet just vanilla SAC
Only real issue is that for some reason I had to comment out the @tf.function build decorators to get SAC to run which doesn't seem correct and might be environment dependent, so maybe those changes specifically shouldn't be merged. I am planning on PR for a Docker container like in Metaworld to ensure that the testing environment is more controllable
maybe this shouldn't be pushed to main, but instead to some new feature branch like "gymnasium"?
Process for building the environment right now (tested Python 3.10):
python3 -m pip install "tensorflow[and-cuda]==2.14"
git clone https://github.com/balloch/Metaworld.git && cd Metaworld && pip install . && cd ..
only really tested with
run_cl.py
so far, and haven't tested the CL methods yet just vanilla SACOnly real issue is that for some reason I had to comment out the
@tf.function
build decorators to get SAC to run which doesn't seem correct and might be environment dependent, so maybe those changes specifically shouldn't be merged. I am planning on PR for a Docker container like in Metaworld to ensure that the testing environment is more controllable