Use Closed-Form Policy to Play BipedalWalker-v3¶

In [1]:
import sys
import logging
import itertools

import numpy as np
np.random.seed(0)
import gym

logging.basicConfig(level=logging.INFO,
        format='%(asctime)s [%(levelname)s] %(message)s',
        stream=sys.stdout, datefmt='%H:%M:%S')
In [2]:
env = gym.make('BipedalWalker-v3')
for key in vars(env):
    logging.info('%s: %s', key, vars(env)[key])
for key in vars(env.spec):
    logging.info('%s: %s', key, vars(env.spec)[key])

00:00:17 [INFO] env: <BipedalWalker<BipedalWalker-v3>>
00:00:17 [INFO] action_space: Box(-1.0, 1.0, (4,), float32)
00:00:17 [INFO] observation_space: Box(-inf, inf, (24,), float32)
00:00:17 [INFO] reward_range: (-inf, inf)
00:00:17 [INFO] metadata: {'render.modes': ['human', 'rgb_array'], 'video.frames_per_second': 50}
00:00:17 [INFO] _max_episode_steps: 1600
00:00:17 [INFO] _elapsed_steps: None
00:00:17 [INFO] id: BipedalWalker-v3
00:00:17 [INFO] entry_point: gym.envs.box2d:BipedalWalker
00:00:17 [INFO] reward_threshold: 300
00:00:17 [INFO] nondeterministic: False
00:00:17 [INFO] max_episode_steps: 1600
00:00:17 [INFO] _kwargs: {}
00:00:17 [INFO] _env_name: BipedalWalker
In [3]:
class ClosedFormAgent:
    def __init__(self, env):
        self.weights = np.array([
            [ 0.9, -0.7,  0.0, -1.4],
            [ 4.3, -1.6, -4.4, -2.0],
            [ 2.4, -4.2, -1.3, -0.1],
            [-3.1, -5.0, -2.0, -3.3],
            [-0.8,  1.4,  1.7,  0.2],
            [-0.7,  0.2, -0.2,  0.1],
            [-0.6, -1.5, -0.6,  0.3],
            [-0.5, -0.3,  0.2,  0.1],
            [ 0.0, -0.1, -0.1,  0.1],
            [ 0.4,  0.8, -1.6, -0.5],
            [-0.4,  0.5, -0.3, -0.4],
            [ 0.3,  2.0,  0.9, -1.6],
            [ 0.0, -0.2,  0.1, -0.3],
            [ 0.1,  0.2, -0.5, -0.3],
            [ 0.7,  0.3,  5.1, -2.4],
            [-0.4, -2.3,  0.3, -4.0],
            [ 0.1, -0.8,  0.3,  2.5],
            [ 0.4, -0.9, -1.8,  0.3],
            [-3.9, -3.5,  2.8,  0.8],
            [ 0.4, -2.8,  0.4,  1.4],
            [-2.2, -2.1, -2.2, -3.2],
            [-2.7, -2.6,  0.3,  0.6],
            [ 2.0,  2.8,  0.0, -0.9],
            [-2.2,  0.6,  4.7, -4.6],
            ])
        self.bias = np.array([3.2, 6.1, -4.0, 7.6])

    def reset(self, mode=None):
        pass

    def step(self, observation, reward, terminated):
        action = np.matmul(observation, self.weights) + self.bias
        return action

    def close(self):
        pass


agent = ClosedFormAgent(env)
In [4]:
def play_episode(env, agent, seed=None, mode=None, render=False):
    observation, _ = env.reset(seed=seed)
    reward, terminated, truncated = 0., False, False
    agent.reset(mode=mode)
    episode_reward, elapsed_steps = 0., 0
    while True:
        action = agent.step(observation, reward, terminated)
        if render:
            env.render()
        if terminated or truncated:
            break
        observation, reward, terminated, truncated, _ = env.step(action)
        episode_reward += reward
        elapsed_steps += 1
    agent.close()
    return episode_reward, elapsed_steps


logging.info('==== test ====')
episode_rewards = []
for episode in range(100):
    episode_reward, elapsed_steps = play_episode(env, agent)
    episode_rewards.append(episode_reward)
    logging.info('test episode %d: reward = %.2f, steps = %d',
            episode, episode_reward, elapsed_steps)
logging.info('average episode reward = %.2f ± %.2f',
        np.mean(episode_rewards), np.std(episode_rewards))

00:00:17 [INFO] ==== test ====
00:00:18 [INFO] test episode 0: reward = 311.54, steps = 1238
00:00:18 [INFO] test episode 1: reward = 311.80, steps = 1246
00:00:19 [INFO] test episode 2: reward = 311.68, steps = 1297
00:00:20 [INFO] test episode 3: reward = 311.93, steps = 1257
00:00:20 [INFO] test episode 4: reward = 312.64, steps = 1279
00:00:21 [INFO] test episode 5: reward = 314.01, steps = 1244
00:00:22 [INFO] test episode 6: reward = 311.10, steps = 1242
00:00:22 [INFO] test episode 7: reward = 312.14, steps = 1286
00:00:23 [INFO] test episode 8: reward = 313.93, steps = 1227
00:00:24 [INFO] test episode 9: reward = 312.31, steps = 1248
00:00:24 [INFO] test episode 10: reward = 313.03, steps = 1243
00:00:25 [INFO] test episode 11: reward = 310.21, steps = 1269
00:00:26 [INFO] test episode 12: reward = 313.31, steps = 1246
00:00:26 [INFO] test episode 13: reward = 311.41, steps = 1246
00:00:27 [INFO] test episode 14: reward = 309.60, steps = 1274
00:00:27 [INFO] test episode 15: reward = 310.79, steps = 1239
00:00:28 [INFO] test episode 16: reward = 312.62, steps = 1240
00:00:29 [INFO] test episode 17: reward = 310.81, steps = 1292
00:00:29 [INFO] test episode 18: reward = 312.69, steps = 1255
00:00:30 [INFO] test episode 19: reward = 311.89, steps = 1254
00:00:31 [INFO] test episode 20: reward = 311.98, steps = 1257
00:00:31 [INFO] test episode 21: reward = 311.88, steps = 1279
00:00:32 [INFO] test episode 22: reward = 312.83, steps = 1257
00:00:32 [INFO] test episode 23: reward = 312.19, steps = 1229
00:00:33 [INFO] test episode 24: reward = 312.23, steps = 1240
00:00:33 [INFO] test episode 25: reward = 311.83, steps = 1243
00:00:34 [INFO] test episode 26: reward = 311.84, steps = 1215
00:00:34 [INFO] test episode 27: reward = 312.74, steps = 1270
00:00:35 [INFO] test episode 28: reward = 312.50, steps = 1243
00:00:36 [INFO] test episode 29: reward = 309.36, steps = 1286
00:00:37 [INFO] test episode 30: reward = 311.43, steps = 1230
00:00:38 [INFO] test episode 31: reward = 313.14, steps = 1227
00:00:38 [INFO] test episode 32: reward = 311.23, steps = 1264
00:00:39 [INFO] test episode 33: reward = 310.01, steps = 1268
00:00:40 [INFO] test episode 34: reward = 312.65, steps = 1261
00:00:41 [INFO] test episode 35: reward = 311.03, steps = 1263
00:00:42 [INFO] test episode 36: reward = 312.31, steps = 1267
00:00:43 [INFO] test episode 37: reward = 311.59, steps = 1297
00:00:44 [INFO] test episode 38: reward = 312.95, steps = 1260
00:00:45 [INFO] test episode 39: reward = 310.43, steps = 1289
00:00:46 [INFO] test episode 40: reward = 310.74, steps = 1273
00:00:46 [INFO] test episode 41: reward = 311.26, steps = 1246
00:00:47 [INFO] test episode 42: reward = 309.09, steps = 1305
00:00:48 [INFO] test episode 43: reward = 312.86, steps = 1251
00:00:48 [INFO] test episode 44: reward = 311.23, steps = 1256
00:00:49 [INFO] test episode 45: reward = 313.63, steps = 1231
00:00:49 [INFO] test episode 46: reward = 314.69, steps = 1200
00:00:50 [INFO] test episode 47: reward = 310.47, steps = 1253
00:00:50 [INFO] test episode 48: reward = 313.85, steps = 1235
00:00:51 [INFO] test episode 49: reward = 310.36, steps = 1271
00:00:52 [INFO] test episode 50: reward = 312.14, steps = 1250
00:00:52 [INFO] test episode 51: reward = 313.43, steps = 1223
00:00:53 [INFO] test episode 52: reward = 312.56, steps = 1239
00:00:53 [INFO] test episode 53: reward = 310.96, steps = 1269
00:00:54 [INFO] test episode 54: reward = 312.64, steps = 1257
00:00:55 [INFO] test episode 55: reward = 313.76, steps = 1239
00:00:55 [INFO] test episode 56: reward = 311.05, steps = 1258
00:00:56 [INFO] test episode 57: reward = 311.65, steps = 1245
00:00:56 [INFO] test episode 58: reward = 311.00, steps = 1249
00:00:57 [INFO] test episode 59: reward = 312.88, steps = 1264
00:00:57 [INFO] test episode 60: reward = 313.16, steps = 1229
00:00:58 [INFO] test episode 61: reward = 310.72, steps = 1259
00:00:58 [INFO] test episode 62: reward = 310.83, steps = 1270
00:00:59 [INFO] test episode 63: reward = 312.81, steps = 1228
00:00:59 [INFO] test episode 64: reward = 310.50, steps = 1248
00:01:00 [INFO] test episode 65: reward = 313.22, steps = 1242
00:01:01 [INFO] test episode 66: reward = 312.31, steps = 1212
00:01:01 [INFO] test episode 67: reward = 312.95, steps = 1232
00:01:02 [INFO] test episode 68: reward = 310.90, steps = 1246
00:01:02 [INFO] test episode 69: reward = 311.22, steps = 1237
00:01:03 [INFO] test episode 70: reward = 312.39, steps = 1249
00:01:03 [INFO] test episode 71: reward = 312.80, steps = 1256
00:01:04 [INFO] test episode 72: reward = 311.77, steps = 1267
00:01:04 [INFO] test episode 73: reward = 310.94, steps = 1253
00:01:05 [INFO] test episode 74: reward = 312.10, steps = 1275
00:01:06 [INFO] test episode 75: reward = 312.31, steps = 1273
00:01:06 [INFO] test episode 76: reward = 313.02, steps = 1248
00:01:07 [INFO] test episode 77: reward = 311.88, steps = 1245
00:01:07 [INFO] test episode 78: reward = 312.70, steps = 1240
00:01:08 [INFO] test episode 79: reward = 313.93, steps = 1225
00:01:08 [INFO] test episode 80: reward = 312.99, steps = 1261
00:01:09 [INFO] test episode 81: reward = 311.63, steps = 1250
00:01:10 [INFO] test episode 82: reward = 313.25, steps = 1245
00:01:10 [INFO] test episode 83: reward = 312.35, steps = 1279
00:01:11 [INFO] test episode 84: reward = 310.08, steps = 1289
00:01:11 [INFO] test episode 85: reward = 309.38, steps = 1273
00:01:12 [INFO] test episode 86: reward = 313.06, steps = 1220
00:01:12 [INFO] test episode 87: reward = 312.29, steps = 1261
00:01:13 [INFO] test episode 88: reward = 312.16, steps = 1264
00:01:14 [INFO] test episode 89: reward = 313.73, steps = 1219
00:01:14 [INFO] test episode 90: reward = 311.74, steps = 1238
00:01:15 [INFO] test episode 91: reward = 312.11, steps = 1242
00:01:16 [INFO] test episode 92: reward = 310.96, steps = 1262
00:01:16 [INFO] test episode 93: reward = 311.33, steps = 1249
00:01:17 [INFO] test episode 94: reward = 309.44, steps = 1285
00:01:18 [INFO] test episode 95: reward = 311.50, steps = 1259
00:01:18 [INFO] test episode 96: reward = 311.90, steps = 1239
00:01:19 [INFO] test episode 97: reward = 311.58, steps = 1258
00:01:19 [INFO] test episode 98: reward = 313.10, steps = 1232
00:01:20 [INFO] test episode 99: reward = 310.88, steps = 1248
00:01:20 [INFO] average episode reward = 311.94 ± 1.15