boringAI/boringAI_rllib_cont_fullwait.py at main · Modjular/boringAI

Repository navigation

History

100

101

102

103

104

105

106

107

108

109

110

111

112

113

114

115

116

117

118

119

120

121

122

123

124

125

126

127

128

129

130

131

132

133

134

135

136

137

138

139

140

141

142

143

144

145

146

147

148

149

150

151

152

153

154

155

156

157

158

159

160

161

162

163

164

165

166

167

168

169

170

171

172

173

174

175

176

177

178

179

180

181

182

183

184

185

186

187

188

189

190

191

192

193

194

195

196

197

198

199

200

201

202

203

204

205

206

207

208

209

210

211

212

213

214

215

216

217

218

219

220

221

222

223

224

225

226

227

228

229

230

231

232

233

234

235

236

237

238

239

240

241

242

243

244

245

246

247

248

249

250

251

252

253

254

255

256

257

258

259

260

261

262

263

264

265

266

267

268

269

270

271

272

273

274

275

276

277

278

279

280

281

282

283

284

285

286

287

288

289

290

291

292

293

294

295

296

297

298

299

300

301

302

303

304

305

306

307

308

309

310

311

312

313

314

315

316

317

318

319

320

321

322

323

324

325

326

327

328

329

330

331

332

333

334

335

336

337

338

339

340

341

342

343

344

345

346

347

348

349

350

351

352

353

354

355

356

357

358

359

360

361

362

363

364

365

366

367

368

369

370

371

372

373

374

375

376

377

378

379

380

381

382

383

384

385

386

387

388

389

390

391

392

393

394

395

396

397

398

399

400

401

402

403

404

405

406

407

408

409

410

411

412

413

414

415

416

417

418

419

420

421

422

423

424

425

426

427

428

429

430

431

432

433

434

435

436

437

438

439

440

441

442

443

444

445

446

447

448

449

450

451

452

453

454

455

456

457

458

459

460

461

462

463

464

465

# Rllib docs: https://docs.ray.io/en/latest/rllib.html

# This AI waits for the block to break before moving on to the next step

try:

from malmo import MalmoPython

except:

import MalmoPython

import sys

import time

import json

import matplotlib.pyplot as plt

import numpy as np

from numpy.random import randint

from collections import defaultdict

import gym, ray

from gym.spaces import Discrete, MultiDiscrete

from ray.rllib.agents import ppo

import random

import datetime

import pandas as pd

INITIAL_REWARD = 0

class BoringAI(gym.Env):

def __init__(self, env_config):

# Static Parameters

self.size = 10

self.reward_density = .1

self.penalty_density = .02

self.obs_size = 1

self.max_episode_steps = 100

self.log_frequency = 5

self.golden_durability = 2

self.start_time = datetime.datetime.now()

print("Init at ", self.start_time)

# BoringAI parameters

self.tunnel_len = 9

self.block_type = ['stone', 'dirt', 'planks','prismarine','clay','log']

self.block_dict = {'stone': 0, 'dirt': 1, 'planks': 2,

'prismarine' :3, 'clay': 4, 'log':5,

'bedrock':6, 'air':6}

self.tool_dict = {0:"diamond pickaxe", 1:"diamond shovel",2:"diamond axe",

3:"golden pickaxe", 4:"golden shovel", 5:"golden axe"}

self.total_tool_type = 3

self.total_tool_material = 2

self.total_block_type = len(self.block_dict.keys()) - 1 #minus 1 due to air and bedrock counting as other

self.golden_durability_dict = defaultdict(lambda:self.golden_durability) # currently only used for golden, takes in index of action

# Rllib Parameters

# self.action_space = Box(-1, 1, shape=(3,), dtype=np.float32)

# self.observation_space = Box(0, 1, shape=(np.prod([2, self.obs_size, self.obs_size]), ), dtype=np.int32)

self.action_space = MultiDiscrete([self.total_tool_type,self.total_tool_material] )

# [pickaxe, shovel, axe] [diamond,golden]

self.observation_space = MultiDiscrete([self.total_block_type,2]); # ([air, dirt, stone, planks], [no attack, attack])

# Malmo Parameters

self.agent_host = MalmoPython.AgentHost()

try:

self.agent_host.parse( sys.argv )

except RuntimeError as e:

print('ERROR:', e)

print(self.agent_host.getUsage())

exit(1)

# DiamondCollector Parameters

self.obs = None

self.allow_break_action = False

self.episode_step = 0

self.episode_return = 0

self.returns = []

self.steps = []

self.initial_reward = INITIAL_REWARD # used per episode to keep track of ticks

self.episode_action_log = defaultdict(lambda:[0,0]) # for each ep, [right uses,total]

self.action_log = defaultdict(lambda:[]) #for each tool, has a list of percentage used correctly per episode

def reset(self):

"""

Resets the environment for the next episode.

Returns

observation: <np.array> flattened initial obseravtion

"""

# Reset Malmo

world_state = self.init_malmo()

# Reset Variables

self.returns.append(self.episode_return)

current_step = self.steps[-1] if len(self.steps) > 0 else 0

self.steps.append(current_step + self.episode_step)

self.episode_return = 0

self.episode_step = 0

self.initial_reward = INITIAL_REWARD

self.golden_durability_dict.clear()

self.golden_durablity_dict = defaultdict(lambda:self.golden_durability)

#self.episode_action_log = defaultdict(lambda:[0,0])#uncomment when you want to have per episode statistics

# Log

if len(self.returns) > 0:

self.log_returns()

if len(self.action_log) >0:

self.log_actions()

# Get Observation

self.obs, self.allow_break_action = self.get_observation(world_state)

plt.close("all")

# return self.obs.flatten()

return [self.obs, 1 if self.allow_break_action else 0]

def step(self, action):

"""

Take an action in the environment and return the results.

Args

action: <np.array> [[action][material]]

[[pickaxe,shovel,axe][diamond,golden]]

Returns

observation: <np.array> flattened array of obseravtion

reward: <int> reward from taking action

done: <bool> indicates terminal state

info: <dict> dictionary of extra information

"""

reward_delta = 0 # calling getWorldState() below loses our reward delta

i_action = action[0]+action[1]*self.total_tool_type

#find hotbar index of the action

if action[1] == 1:

#penalize gold tools and action log counts as incorrect for more than usage

if self.golden_durability_dict[i_action]<1:

reward_delta += 200

else:

if (i_action%self.total_tool_type==self.obs):

self.episode_action_log[i_action][0]+=1

self.golden_durability_dict[i_action]-=1

else:

#penalize if using normal tools with special blocks and action log counts as incorrect

if (self.obs >2 and self.obs<6):

reward_delta +=200

else:

if (i_action%self.total_tool_type==self.obs):

self.episode_action_log[i_action][0]+=1

#action logging

self.episode_action_log[i_action][1] +=1

# Get Action

if self.allow_break_action:

# switch tools, start digging

self.agent_host.sendCommand('move 0');

self.agent_host.sendCommand('hotbar.' + str(i_action + 1) + ' 1')

self.agent_host.sendCommand('attack 1')

# time.sleep(0.8) # Allow steve to break the block

# TODO

# Add a while loop that lets us finish destroying the block

# Or maybe this is a behavior it can learn on its own?

# Otherwise, it might switch randomly, resetting the breaking

while self.allow_break_action:

time.sleep(0.2)

world_state = self.agent_host.getWorldState()

for r in world_state.rewards: # these deltas will be lost otherwise

reward_delta += r.getValue()

_, self.allow_break_action = self.get_observation(world_state)

# just move forward if no block in front

self.agent_host.sendCommand('attack 0')

self.agent_host.sendCommand('move 1');

time.sleep(0.5)

self.episode_step += 1

# Get Observation

world_state = self.agent_host.getWorldState()

for error in world_state.errors:

print("Error:", error.text)

self.obs, self.allow_break_action = self.get_observation(world_state)

# Get Done

done = not world_state.is_mission_running

# Get Reward

reward = 0

for r in world_state.rewards:

reward_delta += r.getValue()

self.initial_reward += reward_delta # increment reward

print("REWARD DELTA(# ticks taken): ", reward_delta)

# NOTE

# "Error: AgentHost::sendCommand : commands connection is not open. Is the mission running?"

# The error above causes <RewardForTimeTaken> to register inconsistently

# To be consistent, we keep track of seblocks/second*scalarlf.initial_reward internally

if done:

#reward is converted to blocks/minute

reward += (int)(self.tunnel_len/self.initial_reward*20*60) # add to reward so it can get passed out

for action in range(self.total_tool_material*self.total_tool_type):

self.action_log[action].append([self.episode_action_log[action][0],self.episode_action_log[action][1]])

self.episode_returns=self.initial_reward

print("END REWARD (blocks/minute)", reward)

self.episode_return += reward

return [self.obs, 1 if self.allow_break_action else 0], reward, done, dict()

def get_mission_xml(self):

tunnel_xml = ''

# Finish Line

for i in range(-5, 6):

if i%2 == 0:

tunnel_xml += "<DrawBlock x='{}' y='1' z='{}' type='coal_block' />".format(i, self.tunnel_len + 1)

else:

tunnel_xml += "<DrawBlock x='{}' y='1' z='{}' type='quartz_block' />".format(i, self.tunnel_len + 1)

# Glass Box

tunnel_xml += "<DrawCuboid x1='-5' x2='5' y1='2' y2='4' z1='1' z2='1' type='glass'/>"

tunnel_xml += "<DrawCuboid x1='-5' x2='-5' y1='2' y2='4' z1='1' z2='{}' type='glass'/>".format(self.tunnel_len)

tunnel_xml += "<DrawCuboid x1='5' x2='5' y1='2' y2='4' z1='1' z2='{}' type='glass'/>".format(self.tunnel_len)

# Draw entrance

#tunnel_xml += "<DrawBlock x='0' y='2' z='1' type='air' />"

tunnel_xml += "<DrawBlock x='0' y='3' z='1' type='air' />"

# NOTE

# Was running into wild results because lots of stone = bad results

# while lots of dirt = quick results

# But it wasn't really learning. Just chance

# Defining equal amounts of each block = more consistency

blocks = ['stone', 'stone', 'prismarine', 'dirt', 'dirt', 'clay',

'planks', 'planks', 'log']

random.shuffle(blocks)

# Blocks to Break

for i in range(0, self.tunnel_len):

tunnel_xml += "<DrawBlock x='0' y='2' z='{}' type='{}' />".format(i + 1, blocks[i])

# tunnel_xml += "<DrawBlock x=\'0\' y=\'3\' z=\'" + str(i) + "\' type=\'" + random_block + "\' />" # uncomment for 2nd row

return '''<?xml version="1.0" encoding="UTF-8" standalone="no" ?>

<About>

<Summary>Tunnel Crawler</Summary>

</About>

<Time>

</Time>

<Weather>clear</Weather>

</ServerInitialConditions>

<DrawingDecorator>''' + \

"<DrawCuboid x1='{}' x2='{}' y1='2' y2='2' z1='{}' z2='{}' type='air'/>".format(-self.size, self.size, -self.size, self.size) + \

"<DrawCuboid x1='{}' x2='{}' y1='1' y2='1' z1='{}' z2='{}' type='bedrock'/>".format(-self.size, self.size, -self.size, self.size) + \

tunnel_xml + \

'''

</DrawingDecorator>

</ServerHandlers>

</ServerSection>

<Name>BoringAI</Name>

</Inventory>

</AgentStart>

</Grid>

</ObservationFromGrid>

</AgentQuitFromTouchingBlockType>

</AgentHandlers>

</AgentSection>

</Mission>'''

def init_malmo(self):

"""

Initialize new malmo mission.

"""

my_mission = MalmoPython.MissionSpec(self.get_mission_xml(), True)

my_mission_record = MalmoPython.MissionRecordSpec()

my_mission.requestVideo(800, 500)

my_mission.setViewpoint(1)

max_retries = 3

my_clients = MalmoPython.ClientPool()

my_clients.add(MalmoPython.ClientInfo('127.0.0.1', 10000)) # add Minecraft machines here as available

for retry in range(max_retries):

try:

self.agent_host.startMission( my_mission, my_clients, my_mission_record, 0, 'DiamondCollector' )

break

except RuntimeError as e:

if retry == max_retries - 1:

print("Error starting mission:", e)

exit(1)

else:

time.sleep(2)

world_state = self.agent_host.getWorldState()

while not world_state.has_mission_begun:

time.sleep(0.1)

world_state = self.agent_host.getWorldState()

for error in world_state.errors:

print("\nError:", error.text)

return world_state

def get_observation(self, world_state):

"""

Use the agent observation API to get the block right in front of it (text)

Args

world_state: <object> current agent world state

Returns

observation: <np.array>

"""

obs = 0

allow_break_action = False

while world_state.is_mission_running:

time.sleep(0.1)

world_state = self.agent_host.getWorldState()

if len(world_state.errors) > 0:

raise AssertionError('Could not load grid.')

if world_state.number_of_observations_since_last_state > 0:

# First we get the json from the observation API

msg = world_state.observations[-1].text

observations = json.loads(msg)

# Get observation

obs = observations['floorAll']

obs = self.block_dict[obs[0]];

# For extra safety, the ground is unbreakable bedrock

allow_break_action = observations['LineOfSight']['type'] != 'bedrock'

break

return obs, allow_break_action

def log_returns(self):

"""

Log the current returns as a graph and text file

Args:

steps (list): list of global steps after each episode

returns (list): list of total return of each episode

"""

box = np.ones(self.log_frequency) / self.log_frequency

returns_smooth = np.convolve(self.returns, box, mode='same')

"""

plt.figure()

plt.style.use('seaborn-darkgrid')

palette = plt.get_cmap('Set1')

plt.clf()

plt.plot(list(range(len(self.returns))), self.returns,marker='',

color=palette(1), linewidth=1, alpha=0.9)

plt.title('BoringAI Continuous Sparse', loc='left', fontsize=12, fontweight=0,

color='orange')

plt.ylabel('Return')

plt.xlabel('Episodes')

plt.savefig('returns.png')

with open('returns.txt', 'w') as f:

for episode, value in enumerate( self.returns):

f.write("{}\t{}\n".format(episode, value))

def log_actions(self):

"""

Log the number of actions taken correctly. e.g. axe to plank

Args:

action_log (dict): dict of array which for each block stores a

number in the corresponding idx

"""

plt.figure()

plt.style.use('seaborn-darkgrid')

palette = plt.get_cmap('Paired')

num=-2

for i_tool in range(self.total_tool_type*self.total_tool_material):

if (num > 3):

num=-1

num+=2

y_values = []

stats=[0,0]

for tmp in self.action_log[i_tool]:

stats[0]+= tmp[0]

stats[1]+= tmp[1]

if stats[1]!=0:

y_values.append((stats[0]/stats[1]))

else:

y_values.append(1)

plt.plot(list(range(len(self.action_log[i_tool]))),y_values,

marker='', color=palette(num), linewidth=1.2, alpha=0.9,

label=self.tool_dict[i_tool])

plt.legend(loc=2, ncol=2)

plt.title("Tool Usage Statistics", loc='left', fontsize=12,

fontweight=0, color='orange')

plt.xlabel("Episode")

plt.ylabel("Percentage Correctly Used")

plt.savefig("toolstats.png")

with open('toolstats.txt', 'w') as f:

for episode in range(len(self.action_log[0])):

f.write("{}\t".format(episode))

for i_tool in range(3):

f.write("{}\t{}\t".format(

self.action_log[i_tool][episode][0],self.action_log[i_tool][episode][1]))

f.write("\n")

if __name__ == '__main__':

ray.init()

trainer = ppo.PPOTrainer(env=BoringAI, config={

'env_config': {}, # No environment parameters to configure

'framework': 'torch', # Use pyotrch instead of tensorflow

'num_gpus': 0, # We aren't using GPUs

'num_workers': 0 # We aren't using parallelism

})

while True:

print(trainer.train())

FilesExpand file tree

boringAI_rllib_cont_fullwait.py

Latest commit

History

boringAI_rllib_cont_fullwait.py

File metadata and controls