def test_infinite_agents():
env = MultiAgentRoundaboutEnv(
{
"map_config": dict(exit_length=20, lane_num=2),
# "use_render": True, "fast": True,
"num_agents": -1,
"delay_done": 50,
"horizon": 50,
"debug": True,
"random_traffic": False
}
)
try:
o = env.reset()
env.seed(100)
env._DEBUG_RANDOM_SEED = 100
max_num = old_num_of_vehicles = len(env.vehicles)
for i in range(1, 1000):
o, r, d, info = env.step({k: [1, 1] for k in env.vehicles})
print(
"{} Current active agents: ".format(i), len(env.vehicles), ". Objects: ",
len(env.agent_manager._object_to_agent)
)
max_num = max(len(env.vehicles), max_num)
# env.render(mode="top_down")
for kkk, iii in info.items():
if d[kkk]:
assert iii["episode_length"] > 1
if d["__all__"]:
o = env.reset()
print("Finish {} steps.".format(i))
finally:
env._DEBUG_RANDOM_SEED = None
env.close()
env = MultiAgentRoundaboutEnv({"num_agents": -1, "delay_done": 0, "horizon": 50, "debug": True})
try:
o = env.reset()
max_num = old_num_of_vehicles = len(env.vehicles)
for i in range(1, 300):
o, r, d, info = env.step({k: [0, 1] for k in env.vehicles})
# print("Current active agents: ", len(env.vehicles),
# ". Objects: ", len(env.agent_manager._object_to_agent))
max_num = max(len(env.vehicles), max_num)
# env.render(mode="top_down")
for kkk, iii in info.items():
if d[kkk]:
assert iii["episode_length"] > 1
if d["__all__"]:
o = env.reset()
print("Finish {} steps.".format(i))
finally:
env.close()
assert max_num > old_num_of_vehicles
def test_ma_roundabout_reward_done_alignment():
# out of road
env = MultiAgentRoundaboutEnv({"horizon": 200, "num_agents": 4, "out_of_road_penalty": 777, "crash_done": False})
try:
_check_spaces_before_reset(env)
obs = env.reset()
_check_spaces_after_reset(env, obs)
assert env.observation_space.contains(obs)
for action in [-1, 1]:
for step in range(5000):
act = {k: [action, 1] for k in env.vehicles.keys()}
o, r, d, i = _act(env, act)
for kkk, ddd in d.items():
if ddd and kkk != "__all__":
assert r[kkk] == -777
assert i[kkk][TerminationState.OUT_OF_ROAD]
# print('{} done passed!'.format(kkk))
for kkk, rrr in r.items():
if rrr == -777:
assert d[kkk]
assert i[kkk][TerminationState.OUT_OF_ROAD]
# print('{} reward passed!'.format(kkk))
if d["__all__"]:
env.reset()
break
finally:
env.close()
# crash
env = MultiAgentRoundaboutEnv(
{
"horizon": 100,
"num_agents": 2,
"crash_vehicle_penalty": 1.7777,
"crash_done": True,
"delay_done": 0,
# "use_render": True,
# "fast": True,
"top_down_camera_initial_z": 160
}
)
# Force the seed here so that the agent1 and agent2 are in same heading! Otherwise they might be in vertical
# heading and cause one of the vehicle raise "out of road" error!
env._DEBUG_RANDOM_SEED = 1
try:
_check_spaces_before_reset(env)
obs = env.reset()
_check_spaces_after_reset(env, obs)
for step in range(5):
act = {k: [0, 0] for k in env.vehicles.keys()}
o, r, d, i = _act(env, act)
env.vehicles["agent0"].set_position(env.vehicles["agent1"].position, height=1.2)
for step in range(5000):
act = {k: [0, 0] for k in env.vehicles.keys()}
o, r, d, i = _act(env, act)
if not any(d.values()):
continue
assert sum(d.values()) == 2
for kkk in ['agent0', 'agent1']:
iii = i[kkk]
assert iii[TerminationState.CRASH_VEHICLE]
assert iii[TerminationState.CRASH]
assert r[kkk] == -1.7777
# for kkk, ddd in d.items():
ddd = d[kkk]
if ddd and kkk != "__all__":
assert r[kkk] == -1.7777
assert i[kkk][TerminationState.CRASH_VEHICLE]
assert i[kkk][TerminationState.CRASH]
# print('{} done passed!'.format(kkk))
# for kkk, rrr in r.items():
rrr = r[kkk]
if rrr == -1.7777:
assert d[kkk]
assert i[kkk][TerminationState.CRASH_VEHICLE]
assert i[kkk][TerminationState.CRASH]
# print('{} reward passed!'.format(kkk))
# assert d["__all__"]
# if d["__all__"]:
break
finally:
env._DEBUG_RANDOM_SEED = None
env.close()
# crash with real fixed vehicle
# crash 2
env = MultiAgentRoundaboutEnv(
{
"map_config": {
"exit_length": 110,
"lane_num": 1
},
# "use_render": True,
# "fast": True,
"horizon": 200,
"num_agents": 40,
"crash_vehicle_penalty": 1.7777,
}
)
try:
_check_spaces_before_reset(env)
obs = env.reset()
_check_spaces_after_reset(env, obs)
for step in range(1):
act = {k: [0, 0] for k in env.vehicles.keys()}
o, r, d, i = _act(env, act)
for v_id, v in env.vehicles.items():
if v_id != "agent0":
v.set_static(True)
for step in range(5000):
act = {k: [0, 1] for k in env.vehicles.keys()}
o, r, d, i = _act(env, act)
for kkk, iii in i.items():
if iii[TerminationState.CRASH]:
assert iii[TerminationState.CRASH_VEHICLE]
if iii[TerminationState.CRASH_VEHICLE]:
assert iii[TerminationState.CRASH]
assert r[kkk] == -1.7777
for kkk, ddd in d.items():
if ddd and kkk != "__all__":
assert i[kkk][TerminationState.OUT_OF_ROAD]
# print('{} done passed!'.format(kkk))
for kkk, rrr in r.items():
if rrr == -1.7777:
# assert d[kkk]
assert i[kkk][TerminationState.CRASH_VEHICLE]
assert i[kkk][TerminationState.CRASH]
# print('{} reward passed!'.format(kkk))
if d["agent0"]:
break
if d["__all__"]:
break
finally:
env.close()
# success
env = MultiAgentRoundaboutEnv(
{
"horizon": 100,
"num_agents": 2,
"success_reward": 999,
"out_of_road_penalty": 555,
"crash_done": True
}
)
try:
_check_spaces_before_reset(env)
obs = env.reset()
_check_spaces_after_reset(env)
env.vehicles["agent0"].set_position(env.vehicles["agent0"].routing_localization.final_lane.end)
assert env.observation_space.contains(obs)
for step in range(5000):
act = {k: [0, 0] for k in env.vehicles.keys()}
o, r, d, i = _act(env, act)
if d["__all__"]:
break
kkk = "agent0"
assert r[kkk] == 999
assert i[kkk][TerminationState.SUCCESS]
assert d[kkk]
kkk = "agent1"
assert r[kkk] != 999
assert not i[kkk][TerminationState.SUCCESS]
assert not d[kkk]
break
finally:
env.close()