Skip to content

Commit

Permalink
Implement new checkpoint system for #25
Browse files Browse the repository at this point in the history
  • Loading branch information
@bzier
bzier committed Nov 15, 2017
1 parent c6791f0 commit d2fbf4c
Show file tree
Hide file tree
Showing 2 changed files with 115 additions and 73 deletions.
170 changes: 115 additions & 55 deletions gym_mupen64plus/envs/MarioKart64/mario_kart_env.py
View file
Original file line number Diff line number Diff line change
Expand Up @@ -13,17 +13,59 @@
class MarioKartEnv(Mupen64PlusEnv):
__metaclass__ = abc.ABCMeta

LAP_COLOR_MAP = {(214, 156, 222): 1,
(198, 140, 198): 2,
(66, 49, 66): 3}
# Indicates the color value of the pixel at point (203, 51)
# This is where the lap number is present in the default HUD
LAP_COLOR_MAP = {(214, 156, 222): 1, # Lap 1
(198, 140, 198): 2, # Lap 2
( 66, 49, 66): 3} # Lap 3

HUD_PROGRESS_COLOR_VALUES = {(000, 000, 255): 1, # Blue: Lap 1
(255, 255, 000): 2, # Yellow: Lap 2
(255, 000, 000): 3} # Red: Lap 3

# Sample 4 pixels for each checkpoint to reduce the
# likelihood of a pixel matching the color by chance
CHECKPOINT_LOCATIONS = [
#### Starting with the upper left corner and moving clockwise ####
[( 64, 36), ( 65, 36), ( 64, 37), ( 65, 37)],

#### Across the top ####
[(194, 36), (195, 36), (194, 37), (195, 37)],
[(324, 36), (325, 36), (324, 37), (325, 37)],
[(454, 36), (455, 36), (454, 37), (455, 37)],

#### Upper right corner ####
[(584, 36), (585, 36), (584, 37), (585, 37)],

#### Down the right side ####
[(584, 138), (585, 138), (584, 139), (585, 139)],
[(584, 240), (585, 240), (584, 241), (585, 241)],
[(584, 342), (585, 342), (584, 343), (585, 343)],

#### Lower right corner is empty; grab a chunk above it and to its left ####
[(584, 443), (585, 443), (583, 444), (583, 445)],

#### Across the bottom ####
[(454, 444), (455, 444), (454, 445), (455, 445)],
[(324, 444), (325, 444), (324, 445), (325, 445)],
[(194, 444), (195, 444), (194, 445), (195, 445)],

#### Lower left corner ####
[( 64, 444), ( 65, 444), ( 64, 445), ( 65, 445)],

#### Up the left side ####
[( 64, 342), ( 65, 342), ( 64, 343), ( 65, 343)],
[( 64, 240), ( 65, 240), ( 64, 241), ( 65, 241)],
[( 64, 138), ( 65, 138), ( 64, 139), ( 65, 139)],
]

DEFAULT_STEP_REWARD = -1
LAP_REWARD = 100
CHECKPOINT_REWARD = 100
END_REWARD = 1000
END_DETECTION_REWARD_REFUND = 215

END_EPISODE_THRESHOLD = 30
END_DETECTION_REWARD_REFUND = END_EPISODE_THRESHOLD - 1

PLAYER_ROW = 0
PLAYER_COL = 0
Expand All @@ -43,7 +85,8 @@ def _reset(self):
self.lap = 1

if self.ENABLE_CHECKPOINTS:
self._checkpoint_tracker = [[False for i in range(len(self.CHECKPOINTS))] for j in range(3)]
self._checkpoint_tracker = [[False for i in range(len(self.CHECKPOINT_LOCATIONS))] for j in range(3)]
self.last_known_ckpt = -1

# Nothing to do on the first call to reset()
if self.reset_count > 0:
Expand All @@ -60,7 +103,6 @@ def _reset(self):
for i in range(77):
self.controller_server.send_controls(ControllerState.NO_OP)


return super(MarioKartEnv, self)._reset()

def _get_reward(self):
Expand All @@ -74,77 +116,94 @@ def _get_reward(self):
# Refund the reward lost in the frames between the race finish and end episode detection
return self.END_DETECTION_REWARD_REFUND + self.END_REWARD
else:
if cur_lap != self.lap:
if cur_lap > self.lap:
self.lap = cur_lap
cprint('Lap %s!' % self.lap, 'red')
return self.LAP_REWARD

elif self.ENABLE_CHECKPOINTS and cur_ckpt > -1 and not self._checkpoint_tracker[self.lap - 1][cur_ckpt]:

# Only allow sequential forward achievement, no backward or skipping allowed.
# e.g. If you hit checkpoint 6, you must have hit all prior checkpoints (1-5)
# on this lap for it to count
if not all(self._checkpoint_tracker[self.lap - 1][:-(len(self.CHECKPOINTS)-cur_ckpt)]):
#cprint('CHECKPOINT hit but not achieved (not all prior points were hit)!', 'red')
return self.DEFAULT_STEP_REWARD

cprint('CHECKPOINT achieved!', 'red')
self._checkpoint_tracker[self.lap - 1][cur_ckpt] = True
cprint('CHECKPOINT achieved!', 'red')
return self.CHECKPOINT_REWARD

else:
return self.DEFAULT_STEP_REWARD

def _get_lap(self):
pix_arr = self.numpy_array
point_a = IMAGE_HELPER.GetPixelColor(pix_arr, 203, 51)
if point_a in self.LAP_COLOR_MAP:
return self.LAP_COLOR_MAP[point_a]
else:
# TODO: What should this do? The pixel is not known, so assume same lap?
return self.lap
# The first checkpoint is the upper left corner. It's value should tell us the lap.
ckpt_val = self._evaluate_checkpoint(self.CHECKPOINT_LOCATIONS[0])

# If it is unknown, assume same lap (character icon is likely covering the corner)
return ckpt_val if ckpt_val != -1 else self.lap

def _get_current_checkpoint(self):
cps = map(self._checkpoint, self.CHECKPOINTS)
if any(cps):
#cprint('--------------------------------------------','red')
#cprint('Checkpoints: %s' % cps, 'yellow')
checkpoint_values = map(self._evaluate_checkpoint, self.CHECKPOINT_LOCATIONS)

checkpoint = np.argmax(cps)
# Check if we have achieved any checkpoints
if any(val > -1 for val in checkpoint_values):

# argmin tells us the first index with the lowest value
index_of_lowest_val = np.argmin(checkpoint_values)

#cprint('Checkpoint: %s' % checkpoint, 'cyan')
if index_of_lowest_val != 0:
# If the argmin is anything but 0, we have achieved
# all the checkpoints up through the prior index
checkpoint = index_of_lowest_val - 1
else:
# If the argmin is at index 0, they are all the same value,
# which means we've hit all the checkpoints for this lap
checkpoint = len(checkpoint_values) - 1

#if self.last_known_ckpt != checkpoint:
# cprint('--------------------------------------------','red')
# cprint('Checkpoints: %s' % checkpoint_values, 'yellow')
# cprint('Checkpoint: %s' % checkpoint, 'cyan')

self.last_known_ckpt = checkpoint
return checkpoint
else:
# We're not at a checkpoint
# We haven't hit any checkpoint yet :(
return -1

def _checkpoint(self, checkpoint_points):
# https://stackoverflow.com/a/3844948
# Efficiently determines if all items in a list are equal by
# counting the occurrences of the first item in the list and
# checking if the count matches the length of the list:
def all_equal(self, some_list):
return some_list.count(some_list[0]) == len(some_list)

def _evaluate_checkpoint(self, checkpoint_points):
pix_arr = self.numpy_array
colored_dots = map(lambda point: IMAGE_HELPER.GetPixelColor(pix_arr, point[0], point[1]),
checkpoint_points)
pixel_means = np.mean(colored_dots, 1)
#print colored_dots
#cprint('Pixel means: %s' % pixel_means, 'cyan')
return any(val < 100 for val in pixel_means)
checkpoint_pixels = map(lambda point: IMAGE_HELPER.GetPixelColor(pix_arr, point[0], point[1]),
checkpoint_points)

#print(checkpoint_pixels)

# If the first pixel is not a valid color, no need to check the other three
if not checkpoint_pixels[0] in self.HUD_PROGRESS_COLOR_VALUES:
return -1
# If the first pixel is good, make sure the other three match
elif not self.all_equal(checkpoint_pixels):
return -1
# If all are good, return the corresponding value
else:
return self.HUD_PROGRESS_COLOR_VALUES[checkpoint_pixels[0]]

def _evaluate_end_state(self):
#cprint('Evaluate End State called!','yellow')
pix_arr = self.numpy_array

upper_left = IMAGE_HELPER.GetPixelColor(pix_arr, 19, 19)
upper_right = IMAGE_HELPER.GetPixelColor(pix_arr, 620, 19)
bottom_left = IMAGE_HELPER.GetPixelColor(pix_arr, 19, 460)
bottom_right = IMAGE_HELPER.GetPixelColor(pix_arr, 620, 460)
point_a = IMAGE_HELPER.GetPixelColor(pix_arr, 203, 51)

if upper_left == upper_right == bottom_left == bottom_right:
if point_a in self.LAP_COLOR_MAP:
self.end_episode_confidence += 1
else:
self.end_episode_confidence = 0

if self.end_episode_confidence > self.END_EPISODE_THRESHOLD:
return True
return True
else:
return False
return False

def _navigate_menu(self):
frame = 0
Expand Down Expand Up @@ -204,11 +263,12 @@ def _navigate_menu(self):
action = ControllerState.JOYSTICK_DOWN
cur_row += 1

# Just as the course loads, change the HUD view
if frame in [263, 265]:
self.controller_server.send_controls(ControllerState.NO_OP, r_cbutton=1)
frame += 1
continue
if self.ENABLE_CHECKPOINTS:
# Just as the course loads, change the HUD view
if frame in [263, 265]:
self.controller_server.send_controls(ControllerState.NO_OP, r_cbutton=1)
frame += 1
continue

if action != ControllerState.NO_OP:
print('Frame ' + str(frame) + ': ' + str(action))
Expand All @@ -218,7 +278,7 @@ def _navigate_menu(self):

def _navigate_post_race_menu(self):
frame = 0
while frame < 138:
while frame < 323:
action = ControllerState.NO_OP

# Post race menu (previous choice selected by default)
Expand All @@ -229,10 +289,10 @@ def _navigate_post_race_menu(self):
# - Replay
# - Save Ghost

# 60 - Times screen
# 75 - Post race menu
# 138 - <Level loaded; turn over control>
if frame in [60, 75]:
# 245 - Times screen
# 260 - Post race menu
# 323 - <Level loaded; turn over control>
if frame in [245, 260]:
action = ControllerState.A_BUTTON

if action != ControllerState.NO_OP:
Expand Down
18 changes: 0 additions & 18 deletions gym_mupen64plus/envs/MarioKart64/track_envs.py
View file
Original file line number Diff line number Diff line change
Expand Up @@ -4,24 +4,6 @@ class MarioKartLuigiRacewayEnv(MarioKartEnv):
MAP_SERIES = 0
MAP_CHOICE = 0

CHECKPOINTS = [
#### Straight-away ####
[(563, 317), (564, 317), (565, 317), (566, 317), (567, 317)],
#### Around the first bend ####
[(563, 285), (564, 285), (565, 285), (566, 285), (567, 285)],
[(540, 259), (540, 260), (540, 261), (540, 262), (540, 263)],
[(516, 286), (517, 286), (518, 286), (519, 286), (520, 286)],
##### The angled sections on etiher side of the tunnel #####
[(553, 321), (554, 321), (554, 320), (555, 320), (555, 319)],
[(547, 370), (548, 371), (549, 372), (550, 373)],
#### They're making another left turn ####
[(526, 397), (527, 397), (528, 397), (529, 397), (530, 397)],
[(546, 412), (546, 413), (546, 414), (546, 415), (546, 416)],
[(562, 398), (563, 398), (564, 398), (565, 398), (566, 398)],
#### Straight-away to the line ####
[(563, 380), (564, 380), (565, 380), (566, 380), (567, 380)]
]

class MarioKartMooMooFarmEnv(MarioKartEnv):
MAP_SERIES = 0
MAP_CHOICE = 1
Expand Down

0 comments on commit d2fbf4c

Please sign in to comment.