| 1 | """A bot's own process, in Overcooked-AI's own interface. The bot file defines make_agent(mdp, mlam), returning an |
| 2 | overcooked_ai_py Agent, or a class Agent; each step it gets the OvercookedState, rebuilt here from the referee's line, |
| 3 | and answers with one of Action.ALL_ACTIONS. mlam (the medium-level action manager, which plans motions) is built when |
| 4 | the bot first asks for it. |
| 5 | python3 -I referee/child.py <bot.py> |
| 6 | """ |
| 7 | import os, sys |
| 8 | |
| 9 | sys.path.insert(0, os.path.join(os.path.dirname(os.path.dirname(os.path.abspath(__file__))), 'kit')) |
| 10 | import numpy as np |
| 11 | from host import load, serve |
| 12 | from overcooked_ai_py.mdp.actions import Action |
| 13 | from overcooked_ai_py.mdp.overcooked_env import OvercookedEnv |
| 14 | from overcooked_ai_py.mdp.overcooked_mdp import OvercookedGridworld, OvercookedState |
| 15 | |
| 16 | make = load(sys.argv[1], 'make_agent', 'Agent') |
| 17 | agent = None |
| 18 | |
| 19 | |
| 20 | def handle(msg): |
| 21 | global agent |
| 22 | if 'episode' in msg: |
| 23 | e = msg['episode'] |
| 24 | np.random.seed(e['seed']) |
| 25 | mdp = OvercookedGridworld.from_layout_name(e['layout']) |
| 26 | env = OvercookedEnv.from_mdp(mdp, horizon=e['horizon'], info_level=0) |
| 27 | agent = make(mdp, env.mlam) if make.__name__ == 'make_agent' else make() |
| 28 | agent.set_agent_index(e['seat']) |
| 29 | agent.set_mdp(mdp) |
| 30 | return 'ready' |
| 31 | a, _ = agent.action(OvercookedState.from_dict(msg['state'])) |
| 32 | return Action.ACTION_TO_INDEX[a] |
| 33 | |
| 34 | |
| 35 | serve(handle) |
shoal