Can't use "lifelong=multitask_skill"

[Valarzz]

Hi, when I'm running this command:

export CUDA_VISIBLE_DEVICES=GPU_ID && \
export MUJOCO_EGL_DEVICE_ID=GPU_ID && \
python lotus/lifelong/main_old.py seed=SEED \
                               benchmark_name=BENCHMARK \
                               policy=bc_transformer_policy \
                               lifelong=multitask_skill \
                               skill_learning.exp_name={SKILL_EXP_NAME} \
                               exp={EXP_NAME} \
                               goal_modality=BUDS

It raises an error saying that there's no multitask_skill algorithm. I check with the registered algorithms REGISTERED_ALGOS, and it prints:

{'sequential', 'multitask', 'er', 'agem', 'ewc', 'packnet', 'singletask', 'subskill', 'metacontroller'}

I wonder if I missed any steps so that I didn't have the multitask_skill algorithm.

[wkwan7]

Hi @Valarzz, there's a typo in this script. We need to use "main.py" for hierarchical skill-based policy training. I have updated README. Thanks!