test_flatland_malfunction.py 26.1 KB
Newer Older
u214892's avatar
u214892 committed
1
import random
2
from typing import Dict, List
u214892's avatar
u214892 committed
3

4
import numpy as np
5
from test_utils import Replay, ReplayConfig, run_replay_config, set_penalties_for_replay
6

7
from flatland.core.env_observation_builder import ObservationBuilder
u214892's avatar
u214892 committed
8
from flatland.core.grid.grid4 import Grid4TransitionsEnum
9
from flatland.core.grid.grid4_utils import get_new_position
u214892's avatar
u214892 committed
10
from flatland.envs.agent_utils import RailAgentStatus
u214892's avatar
u214892 committed
11
from flatland.envs.rail_env import RailEnv, RailEnvActions
12
13
14
from flatland.envs.rail_generators import rail_from_grid_transition_map
from flatland.envs.schedule_generators import random_schedule_generator
from flatland.utils.simple_rail import make_simple_rail2
15
16


17
class SingleAgentNavigationObs(ObservationBuilder):
18
    """
19
    We build a representation vector with 3 binary components, indicating which of the 3 available directions
20
21
22
23
24
25
    for each agent (Left, Forward, Right) lead to the shortest path to its target.
    E.g., if taking the Left branch (if available) is the shortest route to the agent's target, the observation vector
    will be [1, 0, 0].
    """

    def __init__(self):
26
        super().__init__()
27
28

    def reset(self):
29
        pass
30

31
    def get(self, handle: int = 0) -> List[int]:
32
33
        agent = self.env.agents[handle]

u214892's avatar
u214892 committed
34
        if agent.status == RailAgentStatus.READY_TO_DEPART:
u214892's avatar
u214892 committed
35
            agent_virtual_position = agent.initial_position
u214892's avatar
u214892 committed
36
        elif agent.status == RailAgentStatus.ACTIVE:
u214892's avatar
u214892 committed
37
            agent_virtual_position = agent.position
u214892's avatar
u214892 committed
38
        elif agent.status == RailAgentStatus.DONE:
u214892's avatar
u214892 committed
39
            agent_virtual_position = agent.target
u214892's avatar
u214892 committed
40
41
42
        else:
            return None

u214892's avatar
u214892 committed
43
        possible_transitions = self.env.rail.get_transitions(*agent_virtual_position, agent.direction)
44
45
46
47
48
49
50
51
52
53
54
        num_transitions = np.count_nonzero(possible_transitions)

        # Start from the current orientation, and see which transitions are available;
        # organize them as [left, forward, right], relative to the current orientation
        # If only one transition is possible, the forward branch is aligned with it.
        if num_transitions == 1:
            observation = [0, 1, 0]
        else:
            min_distances = []
            for direction in [(agent.direction + i) % 4 for i in range(-1, 2)]:
                if possible_transitions[direction]:
u214892's avatar
u214892 committed
55
                    new_position = get_new_position(agent_virtual_position, direction)
u214892's avatar
u214892 committed
56
57
                    min_distances.append(
                        self.env.distance_map.get()[handle, new_position[0], new_position[1], direction])
58
59
60
61
                else:
                    min_distances.append(np.inf)

            observation = [0, 0, 0]
62
            observation[np.argmin(min_distances)] = 1
63
64
65
66
67

        return observation


def test_malfunction_process():
Erik Nygren's avatar
Erik Nygren committed
68
    # Set fixed malfunction duration for this test
69
    stochastic_data = {'prop_malfunction': 1.,
70
                       'malfunction_rate': 1000,
71
                       'min_duration': 3,
Erik Nygren's avatar
Erik Nygren committed
72
                       'max_duration': 3}
73
74
75
76
77
78
79
80
81
82
83
84

    rail, rail_map = make_simple_rail2()

    env = RailEnv(width=25,
                  height=30,
                  rail_generator=rail_from_grid_transition_map(rail),
                  schedule_generator=random_schedule_generator(),
                  number_of_agents=1,
                  stochastic_data=stochastic_data,  # Malfunction data generator
                  obs_builder_object=SingleAgentNavigationObs()
                  )
    # reset to initialize agents_static
Erik Nygren's avatar
Erik Nygren committed
85
    obs, info = env.reset(False, False, True, random_seed=10)
Erik Nygren's avatar
Erik Nygren committed
86

Erik Nygren's avatar
Erik Nygren committed
87
88
    # Check that a initial duration for malfunction was assigned
    assert env.agents[0].malfunction_data['next_malfunction'] > 0
u214892's avatar
u214892 committed
89
90
    for agent in env.agents:
        agent.status = RailAgentStatus.ACTIVE
Erik Nygren's avatar
Erik Nygren committed
91

92
    agent_halts = 0
Erik Nygren's avatar
Erik Nygren committed
93
94
    total_down_time = 0
    agent_old_position = env.agents[0].position
95
96
97

    # Move target to unreachable position in order to not interfere with test
    env.agents[0].target = (0, 0)
98
99
    for step in range(100):
        actions = {}
u214892's avatar
u214892 committed
100

101
102
103
104
        for i in range(len(obs)):
            actions[i] = np.argmax(obs[i]) + 1

        if step % 5 == 0:
Erik Nygren's avatar
Erik Nygren committed
105
            # Stop the agent and set it to be malfunctioning
106
            env.agents[0].malfunction_data['malfunction'] = -1
Erik Nygren's avatar
Erik Nygren committed
107
            env.agents[0].malfunction_data['next_malfunction'] = 0
108
109
            agent_halts += 1

110
111
        obs, all_rewards, done, _ = env.step(actions)

Erik Nygren's avatar
Erik Nygren committed
112
113
114
115
116
117
        if env.agents[0].malfunction_data['malfunction'] > 0:
            agent_malfunctioning = True
        else:
            agent_malfunctioning = False

        if agent_malfunctioning:
Erik Nygren's avatar
Erik Nygren committed
118
            # Check that agent is not moving while malfunctioning
Erik Nygren's avatar
Erik Nygren committed
119
120
121
122
123
            assert agent_old_position == env.agents[0].position

        agent_old_position = env.agents[0].position
        total_down_time += env.agents[0].malfunction_data['malfunction']

Erik Nygren's avatar
Erik Nygren committed
124
    # Check that the appropriate number of malfunctions is achieved
125
    assert env.agents[0].malfunction_data['nr_malfunctions'] == 20, "Actual {}".format(
u214892's avatar
u214892 committed
126
        env.agents[0].malfunction_data['nr_malfunctions'])
Erik Nygren's avatar
Erik Nygren committed
127

Erik Nygren's avatar
Erik Nygren committed
128
    # Check that 20 stops where performed
129
    assert agent_halts == 20
130

Erik Nygren's avatar
Erik Nygren committed
131
132
    # Check that malfunctioning data was standing around
    assert total_down_time > 0
u214892's avatar
u214892 committed
133
134
135
136
137
138


def test_malfunction_process_statistically():
    """Tests hat malfunctions are produced by stochastic_data!"""
    # Set fixed malfunction duration for this test
    stochastic_data = {'prop_malfunction': 1.,
139
140
141
                       'malfunction_rate': 5,
                       'min_duration': 5,
                       'max_duration': 5}
u214892's avatar
u214892 committed
142

143
144
145
146
147
148
    rail, rail_map = make_simple_rail2()

    env = RailEnv(width=25,
                  height=30,
                  rail_generator=rail_from_grid_transition_map(rail),
                  schedule_generator=random_schedule_generator(),
149
                  number_of_agents=10,
150
151
152
                  stochastic_data=stochastic_data,  # Malfunction data generator
                  obs_builder_object=SingleAgentNavigationObs()
                  )
153

154
    # reset to initialize agents_static
Erik Nygren's avatar
Erik Nygren committed
155
    env.reset(True, True, False, random_seed=10)
156

Erik Nygren's avatar
Erik Nygren committed
157
    env.agents[0].target = (0, 0)
158
    # Next line only for test generation
Erik Nygren's avatar
Erik Nygren committed
159
    #agent_malfunction_list = [[] for i in range(20)]
160
    agent_malfunction_list = [[0, 0, 0, 0, 5, 5, 0, 0, 0, 0], [0, 0, 0, 0, 5, 5, 0, 0, 0, 0], [0, 0, 0, 0, 4, 4, 0, 0, 0, 0],
Erik Nygren's avatar
Erik Nygren committed
161
162
163
164
165
166
     [0, 0, 0, 0, 3, 3, 0, 0, 0, 0], [0, 0, 0, 0, 2, 2, 0, 0, 0, 5], [0, 0, 0, 0, 1, 1, 5, 0, 0, 4],
     [0, 0, 0, 5, 0, 0, 4, 5, 0, 3], [5, 0, 0, 4, 0, 0, 3, 4, 0, 2], [4, 5, 0, 3, 5, 5, 2, 3, 5, 1],
     [3, 4, 0, 2, 4, 4, 1, 2, 4, 0], [2, 3, 5, 1, 3, 3, 0, 1, 3, 0], [1, 2, 4, 0, 2, 2, 0, 0, 2, 0],
     [0, 1, 3, 0, 1, 1, 5, 0, 1, 0], [0, 0, 2, 0, 0, 0, 4, 0, 0, 0], [5, 0, 1, 0, 0, 0, 3, 5, 0, 5],
     [4, 0, 0, 0, 5, 0, 2, 4, 0, 4], [3, 0, 0, 0, 4, 0, 1, 3, 5, 3], [2, 0, 0, 0, 3, 0, 0, 2, 4, 2],
     [1, 0, 5, 5, 2, 0, 0, 1, 3, 1], [0, 5, 4, 4, 1, 0, 5, 0, 2, 0]]
167

Erik Nygren's avatar
Erik Nygren committed
168
    for step in range(20):
169
        action_dict: Dict[int, RailEnvActions] = {}
170
        for agent_idx in range(env.get_num_agents()):
u214892's avatar
u214892 committed
171
            # We randomly select an action
172
173
            action_dict[agent_idx] = RailEnvActions(np.random.randint(4))
            # For generating tests only:
Erik Nygren's avatar
Erik Nygren committed
174
            # agent_malfunction_list[step].append(env.agents[agent_idx].malfunction_data['malfunction'])
175
            assert env.agents[agent_idx].malfunction_data['malfunction'] == agent_malfunction_list[step][agent_idx]
u214892's avatar
u214892 committed
176
        env.step(action_dict)
177
178
    # For generating test onlz
    #print(agent_malfunction_list)
179

u214892's avatar
u214892 committed
180

181
def test_malfunction_before_entry():
182
    """Tests that malfunctions are working properlz for agents before entering the environment!"""
183
184
    # Set fixed malfunction duration for this test
    stochastic_data = {'prop_malfunction': 1.,
185
                       'malfunction_rate': 5,
186
187
188
189
190
191
192
193
                       'min_duration': 10,
                       'max_duration': 10}

    rail, rail_map = make_simple_rail2()

    env = RailEnv(width=25,
                  height=30,
                  rail_generator=rail_from_grid_transition_map(rail),
Erik Nygren's avatar
Erik Nygren committed
194
195
196
                  schedule_generator=random_schedule_generator(seed=2),  # seed 12
                  number_of_agents=10,
                  random_seed=1,
197
198
199
                  stochastic_data=stochastic_data,  # Malfunction data generator
                  )
    # reset to initialize agents_static
Erik Nygren's avatar
Erik Nygren committed
200
    env.reset(False, False, False, random_seed=10)
201

202
203
204
205
206
207
208
    # Test initial malfunction values for all agents
    # we want some agents to be malfuncitoning already and some to be working
    # we want different next_malfunction values for the agents
    assert env.agents[0].malfunction_data['malfunction'] == 0
    assert env.agents[1].malfunction_data['malfunction'] == 0
    assert env.agents[2].malfunction_data['malfunction'] == 0
    assert env.agents[3].malfunction_data['malfunction'] == 0
209
210
    assert env.agents[4].malfunction_data['malfunction'] == 10
    assert env.agents[5].malfunction_data['malfunction'] == 10
211
212
213
214
    assert env.agents[6].malfunction_data['malfunction'] == 0
    assert env.agents[7].malfunction_data['malfunction'] == 0
    assert env.agents[8].malfunction_data['malfunction'] == 0
    assert env.agents[9].malfunction_data['malfunction'] == 0
215

216
217
218
219
220
def test_next_malfunction_counter():
    """
    Test that the next malfunction occurs when desired
    Returns
    -------
221

222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
    """
    # Set fixed malfunction duration for this test

    rail, rail_map = make_simple_rail2()
    action_dict: Dict[int, RailEnvActions] = {}

    env = RailEnv(width=25,
                  height=30,
                  rail_generator=rail_from_grid_transition_map(rail),
                  schedule_generator=random_schedule_generator(seed=2),  # seed 12
                  number_of_agents=1,
                  random_seed=1,
                  )
    # reset to initialize agents_static
    env.reset(False, False, activate_agents=True, random_seed=10)
    env.agents[0].malfunction_data['next_malfunction'] = 5
    env.agents[0].malfunction_data['malfunction_rate'] = 5
    env.agents[0].malfunction_data['malfunction'] = 0
    env.agents[0].target =(0, 0), #Move the target out of range
    print(env.agents[0].position, env.agents[0].malfunction_data['next_malfunction'])

    for time_step in range(1, 6):
        # Move in the env
245
        env.step(action_dict)
246

247
248
249
        # Check that next_step decreases as expected
        assert env.agents[0].malfunction_data['next_malfunction'] == 5 - time_step

250

251

252
def test_initial_malfunction():
u214892's avatar
u214892 committed
253
    stochastic_data = {'prop_malfunction': 1.,  # Percentage of defective agents
254
                       'malfunction_rate': 100,  # Rate of malfunction occurence
u214892's avatar
u214892 committed
255
256
257
258
                       'min_duration': 2,  # Minimal duration of malfunction
                       'max_duration': 5  # Max duration of malfunction
                       }

259
260
    rail, rail_map = make_simple_rail2()

u214892's avatar
u214892 committed
261
262
    env = RailEnv(width=25,
                  height=30,
263
                  rail_generator=rail_from_grid_transition_map(rail),
264
                  schedule_generator=random_schedule_generator(seed=10),
u214892's avatar
u214892 committed
265
266
                  number_of_agents=1,
                  stochastic_data=stochastic_data,  # Malfunction data generator
267
                  obs_builder_object=SingleAgentNavigationObs()
u214892's avatar
u214892 committed
268
                  )
269
    # reset to initialize agents_static
Erik Nygren's avatar
Erik Nygren committed
270
    env.reset(False, False, True, random_seed=10)
271
    print(env.agents[0].malfunction_data)
Erik Nygren's avatar
Erik Nygren committed
272
    env.agents[0].target = (0, 5)
273
    set_penalties_for_replay(env)
274
275
276
    replay_config = ReplayConfig(
        replay=[
            Replay(
277
                position=(3, 2),
278
279
280
281
282
283
284
                direction=Grid4TransitionsEnum.EAST,
                action=RailEnvActions.MOVE_FORWARD,
                set_malfunction=3,
                malfunction=3,
                reward=env.step_penalty  # full step penalty when malfunctioning
            ),
            Replay(
285
                position=(3, 2),
286
287
288
289
290
291
292
293
                direction=Grid4TransitionsEnum.EAST,
                action=RailEnvActions.MOVE_FORWARD,
                malfunction=2,
                reward=env.step_penalty  # full step penalty when malfunctioning
            ),
            # malfunction stops in the next step and we're still at the beginning of the cell
            # --> if we take action MOVE_FORWARD, agent should restart and move to the next cell
            Replay(
294
                position=(3, 2),
295
296
297
                direction=Grid4TransitionsEnum.EAST,
                action=RailEnvActions.MOVE_FORWARD,
                malfunction=1,
298
299
300
                reward= env.step_penalty * 1.0

            ),# malfunctioning ends: starting and running at speed 1.0
301
            Replay(
302
                position=(3, 2),
303
                direction=Grid4TransitionsEnum.EAST,
304
305
                action=RailEnvActions.MOVE_FORWARD,
                malfunction=0,
306
                reward=env.start_penalty +env.step_penalty * 1.0  # running at speed 1.0
307
308
            ),
            Replay(
309
                position=(3, 3),
310
                direction=Grid4TransitionsEnum.EAST,
311
312
313
314
315
316
                action=RailEnvActions.MOVE_FORWARD,
                malfunction=0,
                reward=env.step_penalty * 1.0  # running at speed 1.0
            )
        ],
        speed=env.agents[0].speed_data['speed'],
u214892's avatar
u214892 committed
317
        target=env.agents[0].target,
318
        initial_position=(3, 2),
u214892's avatar
u214892 committed
319
        initial_direction=Grid4TransitionsEnum.EAST,
320
    )
321
    run_replay_config(env, [replay_config])
322
323
324


def test_initial_malfunction_stop_moving():
325
326
327
328
329
330
    stochastic_data = {'prop_malfunction': 1.,  # Percentage of defective agents
                       'malfunction_rate': 70,  # Rate of malfunction occurence
                       'min_duration': 2,  # Minimal duration of malfunction
                       'max_duration': 5  # Max duration of malfunction
                       }

331
    rail, rail_map = make_simple_rail2()
332
333
334

    env = RailEnv(width=25,
                  height=30,
335
336
                  rail_generator=rail_from_grid_transition_map(rail),
                  schedule_generator=random_schedule_generator(),
337
338
                  number_of_agents=1,
                  stochastic_data=stochastic_data,  # Malfunction data generator
339
                  obs_builder_object=SingleAgentNavigationObs()
340
                  )
341
    env.reset()
342
343
344

    print(env.agents[0].initial_position, env.agents[0].direction, env.agents[0].position, env.agents[0].status)

345
    set_penalties_for_replay(env)
346
347
348
    replay_config = ReplayConfig(
        replay=[
            Replay(
u214892's avatar
u214892 committed
349
                position=None,
350
                direction=Grid4TransitionsEnum.EAST,
u214892's avatar
u214892 committed
351
                action=RailEnvActions.MOVE_FORWARD,
352
353
                set_malfunction=3,
                malfunction=3,
u214892's avatar
u214892 committed
354
355
                reward=env.step_penalty,  # full step penalty when stopped
                status=RailAgentStatus.READY_TO_DEPART
356
357
            ),
            Replay(
358
                position=(3, 2),
359
360
                direction=Grid4TransitionsEnum.EAST,
                action=RailEnvActions.DO_NOTHING,
361
                malfunction=3,
u214892's avatar
u214892 committed
362
363
                reward=env.step_penalty,  # full step penalty when stopped
                status=RailAgentStatus.ACTIVE
364
365
366
367
368
            ),
            # malfunction stops in the next step and we're still at the beginning of the cell
            # --> if we take action STOP_MOVING, agent should restart without moving
            #
            Replay(
369
                position=(3, 2),
370
371
                direction=Grid4TransitionsEnum.EAST,
                action=RailEnvActions.STOP_MOVING,
372
                malfunction=2,
u214892's avatar
u214892 committed
373
374
                reward=env.step_penalty,  # full step penalty while stopped
                status=RailAgentStatus.ACTIVE
375
376
377
            ),
            # we have stopped and do nothing --> should stand still
            Replay(
378
                position=(3, 2),
379
380
                direction=Grid4TransitionsEnum.EAST,
                action=RailEnvActions.DO_NOTHING,
381
                malfunction=1,
u214892's avatar
u214892 committed
382
383
                reward=env.step_penalty,  # full step penalty while stopped
                status=RailAgentStatus.ACTIVE
384
385
386
            ),
            # we start to move forward --> should go to next cell now
            Replay(
387
                position=(3, 2),
388
389
390
                direction=Grid4TransitionsEnum.EAST,
                action=RailEnvActions.MOVE_FORWARD,
                malfunction=0,
u214892's avatar
u214892 committed
391
392
                reward=env.start_penalty + env.step_penalty * 1.0,  # full step penalty while stopped
                status=RailAgentStatus.ACTIVE
393
394
            ),
            Replay(
395
                position=(3, 3),
396
                direction=Grid4TransitionsEnum.EAST,
397
398
                action=RailEnvActions.MOVE_FORWARD,
                malfunction=0,
u214892's avatar
u214892 committed
399
400
                reward=env.step_penalty * 1.0,  # full step penalty while stopped
                status=RailAgentStatus.ACTIVE
401
402
403
            )
        ],
        speed=env.agents[0].speed_data['speed'],
u214892's avatar
u214892 committed
404
        target=env.agents[0].target,
405
        initial_position=(3, 2),
u214892's avatar
u214892 committed
406
        initial_direction=Grid4TransitionsEnum.EAST,
407
    )
408
409

    run_replay_config(env, [replay_config], activate_agents=False)
410
411


412
def test_initial_malfunction_do_nothing():
413
414
415
416
417
418
419
420
421
    random.seed(0)
    np.random.seed(0)

    stochastic_data = {'prop_malfunction': 1.,  # Percentage of defective agents
                       'malfunction_rate': 70,  # Rate of malfunction occurence
                       'min_duration': 2,  # Minimal duration of malfunction
                       'max_duration': 5  # Max duration of malfunction
                       }

422
423
    rail, rail_map = make_simple_rail2()

424
425
    env = RailEnv(width=25,
                  height=30,
426
427
                  rail_generator=rail_from_grid_transition_map(rail),
                  schedule_generator=random_schedule_generator(),
428
429
430
                  number_of_agents=1,
                  stochastic_data=stochastic_data,  # Malfunction data generator
                  )
431
432
    # reset to initialize agents_static
    env.reset()
433
    set_penalties_for_replay(env)
434
    replay_config = ReplayConfig(
u214892's avatar
u214892 committed
435
436
437
438
439
440
441
442
443
444
        replay=[
            Replay(
                position=None,
                direction=Grid4TransitionsEnum.EAST,
                action=RailEnvActions.MOVE_FORWARD,
                set_malfunction=3,
                malfunction=3,
                reward=env.step_penalty,  # full step penalty while malfunctioning
                status=RailAgentStatus.READY_TO_DEPART
            ),
445
            Replay(
446
                position=(3, 2),
447
448
                direction=Grid4TransitionsEnum.EAST,
                action=RailEnvActions.DO_NOTHING,
449
                malfunction=3,
u214892's avatar
u214892 committed
450
451
                reward=env.step_penalty,  # full step penalty while malfunctioning
                status=RailAgentStatus.ACTIVE
452
453
454
455
456
            ),
            # malfunction stops in the next step and we're still at the beginning of the cell
            # --> if we take action DO_NOTHING, agent should restart without moving
            #
            Replay(
457
                position=(3, 2),
458
459
                direction=Grid4TransitionsEnum.EAST,
                action=RailEnvActions.DO_NOTHING,
460
                malfunction=2,
u214892's avatar
u214892 committed
461
462
                reward=env.step_penalty,  # full step penalty while stopped
                status=RailAgentStatus.ACTIVE
463
464
465
            ),
            # we haven't started moving yet --> stay here
            Replay(
466
                position=(3, 2),
467
468
                direction=Grid4TransitionsEnum.EAST,
                action=RailEnvActions.DO_NOTHING,
469
                malfunction=1,
u214892's avatar
u214892 committed
470
471
                reward=env.step_penalty,  # full step penalty while stopped
                status=RailAgentStatus.ACTIVE
472
            ),
473

474
            Replay(
475
                position=(3, 2),
476
477
478
                direction=Grid4TransitionsEnum.EAST,
                action=RailEnvActions.MOVE_FORWARD,
                malfunction=0,
u214892's avatar
u214892 committed
479
480
                reward=env.start_penalty + env.step_penalty * 1.0,  # start penalty + step penalty for speed 1.0
                status=RailAgentStatus.ACTIVE
481
            ),  # we start to move forward --> should go to next cell now
482
            Replay(
483
                position=(3, 3),
484
                direction=Grid4TransitionsEnum.EAST,
485
486
                action=RailEnvActions.MOVE_FORWARD,
                malfunction=0,
u214892's avatar
u214892 committed
487
488
                reward=env.step_penalty * 1.0,  # step penalty for speed 1.0
                status=RailAgentStatus.ACTIVE
489
490
491
            )
        ],
        speed=env.agents[0].speed_data['speed'],
u214892's avatar
u214892 committed
492
        target=env.agents[0].target,
493
        initial_position=(3, 2),
u214892's avatar
u214892 committed
494
        initial_direction=Grid4TransitionsEnum.EAST,
495
    )
496
    run_replay_config(env, [replay_config], activate_agents=False)
497
498
499
500
501
502
503


def test_initial_nextmalfunction_not_below_zero():
    random.seed(0)
    np.random.seed(0)

    stochastic_data = {'prop_malfunction': 1.,  # Percentage of defective agents
504
505
                       'malfunction_rate': 70,  # Rate of malfunction occurence
                       'min_duration': 2,  # Minimal duration of malfunction
506
507
508
                       'max_duration': 5  # Max duration of malfunction
                       }

509
    rail, rail_map = make_simple_rail2()
510
511
512

    env = RailEnv(width=25,
                  height=30,
513
514
                  rail_generator=rail_from_grid_transition_map(rail),
                  schedule_generator=random_schedule_generator(),
515
516
                  number_of_agents=1,
                  stochastic_data=stochastic_data,  # Malfunction data generator
517
                  obs_builder_object=SingleAgentNavigationObs()
518
                  )
519
520
    # reset to initialize agents_static
    env.reset()
521
522
523
524
525
    agent = env.agents[0]
    env.step({})
    # was next_malfunction was -1 befor the bugfix https://gitlab.aicrowd.com/flatland/flatland/issues/186
    assert agent.malfunction_data['next_malfunction'] >= 0, \
        "next_malfunction should be >=0, found {}".format(agent.malfunction_data['next_malfunction'])
Erik Nygren's avatar
Erik Nygren committed
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545


def tests_random_interference_from_outside():
    """Tests that malfunctions are produced by stochastic_data!"""
    # Set fixed malfunction duration for this test
    stochastic_data = {'prop_malfunction': 1.,
                       'malfunction_rate': 1,
                       'min_duration': 10,
                       'max_duration': 10}

    rail, rail_map = make_simple_rail2()

    env = RailEnv(width=25,
                  height=30,
                  rail_generator=rail_from_grid_transition_map(rail),
                  schedule_generator=random_schedule_generator(seed=2),  # seed 12
                  number_of_agents=1,
                  random_seed=1,
                  stochastic_data=stochastic_data,  # Malfunction data generator
                  )
546
    env.reset()
Erik Nygren's avatar
Erik Nygren committed
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
    # reset to initialize agents_static
    env.agents[0].speed_data['speed'] = 0.33
    env.agents[0].initial_position = (3, 0)
    env.agents[0].target = (3, 9)
    env.reset(False, False, False)
    env_data = []

    for step in range(200):
        action_dict: Dict[int, RailEnvActions] = {}
        for agent in env.agents:
            # We randomly select an action
            action_dict[agent.handle] = RailEnvActions(2)

        _, reward, _, _ = env.step(action_dict)
        # Append the rewards of the first trial
Erik Nygren's avatar
Erik Nygren committed
562
        env_data.append((reward[0], env.agents[0].position))
Erik Nygren's avatar
Erik Nygren committed
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
        assert reward[0] == env_data[step][0]
        assert env.agents[0].position == env_data[step][1]
    # Run the same test as above but with an external random generator running
    # Check that the reward stays the same

    rail, rail_map = make_simple_rail2()
    random.seed(47)
    np.random.seed(1234)
    env = RailEnv(width=25,
                  height=30,
                  rail_generator=rail_from_grid_transition_map(rail),
                  schedule_generator=random_schedule_generator(seed=2),  # seed 12
                  number_of_agents=1,
                  random_seed=1,
                  stochastic_data=stochastic_data,  # Malfunction data generator
                  )
579
    env.reset()
Erik Nygren's avatar
Erik Nygren committed
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
    # reset to initialize agents_static
    env.agents[0].speed_data['speed'] = 0.33
    env.agents[0].initial_position = (3, 0)
    env.agents[0].target = (3, 9)
    env.reset(False, False, False)

    # Print for test generation
    dummy_list = [1, 2, 6, 7, 8, 9, 4, 5, 4]
    for step in range(200):
        action_dict: Dict[int, RailEnvActions] = {}
        for agent in env.agents:
            # We randomly select an action
            action_dict[agent.handle] = RailEnvActions(2)

            # Do dummy random number generations
Erik Nygren's avatar
Erik Nygren committed
595
596
            random.shuffle(dummy_list)
            np.random.rand()
Erik Nygren's avatar
Erik Nygren committed
597
598
599
600

        _, reward, _, _ = env.step(action_dict)
        assert reward[0] == env_data[step][0]
        assert env.agents[0].position == env_data[step][1]
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642


def test_last_malfunction_step():
    """
    Test to check that agent moves when it is not malfunctioning

    """

    # Set fixed malfunction duration for this test
    stochastic_data = {'prop_malfunction': 1.,
                       'malfunction_rate': 5,
                       'min_duration': 4,
                       'max_duration': 4}

    rail, rail_map = make_simple_rail2()

    env = RailEnv(width=25,
                  height=30,
                  rail_generator=rail_from_grid_transition_map(rail),
                  schedule_generator=random_schedule_generator(seed=2),  # seed 12
                  number_of_agents=1,
                  random_seed=1,
                  stochastic_data=stochastic_data,  # Malfunction data generator
                  )
    env.reset()
    # reset to initialize agents_static
    env.agents[0].speed_data['speed'] = 0.33
    env.agents_static[0].target = (0, 0)

    env.reset(False, False, True)
    # Force malfunction to be off at beginning and next malfunction to happen in 2 steps
    env.agents[0].malfunction_data['next_malfunction'] = 2
    env.agents[0].malfunction_data['malfunction'] = 0
    env_data = []

    for step in range(20):
        action_dict: Dict[int, RailEnvActions] = {}
        for agent in env.agents:
            # Go forward all the time
            action_dict[agent.handle] = RailEnvActions(2)

        # Check if the agent is still allowed to move in this step
643
        if env.agents[0].malfunction_data['malfunction'] > 0 or env.agents[0].malfunction_data['next_malfunction'] < 1:
644
645
646
647
648
649
650
651
652
653
654
655
656
657
            agent_can_move = False
        else:
            agent_can_move = True

        # Store the position before and after the step
        pre_position = env.agents[0].speed_data['position_fraction']
        _, reward, _, _ = env.step(action_dict)
        post_position = env.agents[0].speed_data['position_fraction']

        # Assert that the agent moved while it was still allowed
        if agent_can_move:
            assert pre_position != post_position
        else:
            assert post_position == pre_position