ActionEmbedding

This is the code repository for the paper Learning Action - Transferable Policy with Action Embedding.

Experiments

The code provide two sets of experiments

• Gridworld
• InvertedPendulum (need mujoco license)

There needs some modifications for tasks without state embedding

• sac.py
  • Line 40-57
• agent.py
  • Line 64-68

Run

python agent.py
        -i identifier
	-t task_id
			
	-transfer 0/1
	-ckpt_path path
	-ckpt_step step
	-source_t source_task_id