search

schroederdewitt / multiagent_mujoco

Benchmark for Continuous Multi-Agent Robotic Control, based on OpenAI's Mujoco Gym environments.
Apache License 2.0
330 stars 34 forks source link

Bug for Humanoid-v2 #5

Closed parachutel closed 4 years ago

parachutel commented 4 years ago

For 2-Agent Humanoid-v2, the action_space defined in MujocoMulti.__init__ is (Box(8,), Box(8,)), with a flat dim of 16. However, the underlying original mujoco env MujocoMulti.wrapped_env has an action space of Box(17,). There is a mismatch in flat action dims.

schroederdewitt commented 4 years ago

Dear @parachutel, many thanks for raising this issue. I have just pushed a fix for scenarios with unequal agent action dimensions. Please note that the config for Humanoid/Humanoid-Standup has changed slightly:

2-Agent Humanoid

env_args.scenario="Humanoid-v2"
env_args.agent_conf="9|8"
env_args.agent_obsk=1

2-Agent HumanoidStandup


env_args.scenario="HumanoidStandup-v2"
env_args.agent_conf="9|8"
env_args.agent_obsk=1