Closed belerico closed 1 year ago
I've trained a Walker-Walk agent with the following:
PYOPENGL_PLATFORM="" MUJOCO_GL=osmesa CUDA_VISIBLE_DEVICES="2,3" lightning run model --devices=2 --accelerator=gpu sheeprl.py sac_pixel_continuous --env_id=dmc_walker_walk --action_repeat=2 --capture_video --sample_next_obs --checkpoint_every=50000 --seed=1
obtaining:
This PR implements a SAC algorithm working with images. It is inspired by https://arxiv.org/abs/1910.01741