参数设置 - DailinH/MAgent GitHub Wiki

默认为False。设置方式例如：cfg.set({"minimap_mode": True})

• food_mode: dead agent will leave food in the map

• turn_mode: has two more actions -- turn left and turn right

• minimap_mode: add minimap into observation

  (见GridWorld.cc设置) 在minimap情形下，暂存视野范围内内容，暂存各group目前元素数量，然后各agent间共享信息

• goal_mode: [deprecated!] every agent has a specific goal

• large_map_mode:

• mean_mode: [deprecated!]

• height: int, height of agent body
• width: int, width of agent body
• speed: float, maximum speed, i.e. the radius of move circle of the agent
• hp: float, maximum health point of the agent
• view_range: gw.CircleRange or gw.SectorRange (目前似乎不支持gw.SectorRange，训练结果差)
• damage: float, attack damage(给对方造成的damage)
• step_recover: float, step recover of health point (can be negative)
• kill_supply: float, the hp gain when kill this type of agents
• step_reward: float, reward get in every step
• kill_reward: float, reward gain when kill this type of agent
• dead_penalty: float, reward get when dead
• attack_penalty: float, reward get when perform an attack (this is used to make agents do not attack blank grid)

	predator	prey
size	2*2	1*1
hp	1	1
speed	1	1.5
view_range	半径为5px	半径为4px
attack_range	半径为4px	半径为0px（无法attack）
attack_penalty	-0.2	无

python/magent/builtin/config/pursuit.py

	tiger	deer
size	1*1	1*1
hp	10	5
speed	1	1
view_range	半径为4px	半径为1px
attack_range	半径为1px	半径为0px（无法attack）
kill_supply(????)	0	8
attack_penalty	-0.1	无
step_reward	1
damage	3	0
step_recover	-0.5	0.2

python/magent/builtin/config/forest.py

基本同forest模型，两组tiger，一组deer

python/magent/builtin/config/double_attack.py

两组 small 相互攻击

	small
size	1*1
hp	10
speed	2
view_range	radius=5px
attack_range	radius=1.5px
damage	2
step_recover	0.1
step_reward	-0.005
kill_reward	5
dead_penalty	-0.1
attack_penalty	-0.1

python/magent/builtin/config/double_attack.py