File size: 10,899 Bytes
6dd47af
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
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
#!/usr/bin/env python3
"""Interactive cartpole control via OpenEnv.

This example demonstrates using the dm_control OpenEnv client with
the cartpole environment. Use arrow keys to control the cart.

Controls:
    LEFT/RIGHT arrows: Apply force to move cart
    R: Reset environment
    ESC or Q: Quit

Requirements:
    pip install pygame

Usage:
    1. Start the server: uvicorn server.app:app --host 0.0.0.0 --port 8000
    2. Run this script: python examples/cartpole_control.py

    For visual mode (requires working MuJoCo rendering):
        python examples/cartpole_control.py --visual
"""

import argparse
import random
import sys
from pathlib import Path

# Add parent directory to path for imports
sys.path.insert(0, str(Path(__file__).parent.parent))

from client import DMControlEnv
from models import DMControlAction


def run_headless(env: DMControlEnv, task: str = "balance", max_steps: int = 500):
    """Run cartpole control in headless mode."""
    print("\n=== Headless Mode (OpenEnv Step/Observation Pattern) ===")
    print("This mode demonstrates the OpenEnv API with the cartpole.\n")

    # Reset environment using OpenEnv pattern
    result = env.reset(domain_name="cartpole", task_name=task)
    print(f"Initial observations: {list(result.observation.observations.keys())}")
    print(f"  position: {result.observation.observations.get('position', [])}")
    print(f"  velocity: {result.observation.observations.get('velocity', [])}")

    total_reward = 0.0
    step_count = 0

    print("\nRunning with random actions to demonstrate step/observation pattern...\n")

    while not result.done and step_count < max_steps:
        # Random action in [-1, 1]
        action_value = random.uniform(-1.0, 1.0)

        # Step the environment using OpenEnv pattern
        action = DMControlAction(values=[action_value])
        result = env.step(action)

        # Access observation and reward from result
        total_reward += result.reward or 0.0
        step_count += 1

        # Print progress periodically
        if step_count % 50 == 0:
            pos = result.observation.observations.get("position", [])
            vel = result.observation.observations.get("velocity", [])
            print(
                f"Step {step_count}: reward={result.reward:.3f}, "
                f"total={total_reward:.2f}, done={result.done}"
            )
            print(f"  position={pos}, velocity={vel}")

    print(f"\nEpisode finished: {step_count} steps, total reward: {total_reward:.2f}")


def run_interactive(env: DMControlEnv, task: str = "balance"):
    """Run interactive control with keyboard input via pygame."""
    import pygame

    print("\n=== Interactive Mode (OpenEnv Step/Observation Pattern) ===")
    print("Use LEFT/RIGHT arrows to control cart, R to reset, ESC to quit.\n")

    # Reset environment using OpenEnv pattern
    result = env.reset(domain_name="cartpole", task_name=task)
    print(f"Initial observations: {list(result.observation.observations.keys())}")

    # Initialize pygame for keyboard input (minimal window)
    pygame.init()
    screen = pygame.display.set_mode((400, 100))
    pygame.display.set_caption("Cartpole Control - Arrow keys to move, R to reset")
    clock = pygame.time.Clock()

    # Font for display
    font = pygame.font.Font(None, 24)

    running = True
    total_reward = 0.0
    step_count = 0

    print("\nControls:")
    print("  LEFT/RIGHT arrows: Move cart")
    print("  R: Reset environment")
    print("  ESC or Q: Quit\n")

    while running:
        # Handle events
        for event in pygame.event.get():
            if event.type == pygame.QUIT:
                running = False
            elif event.type == pygame.KEYDOWN:
                if event.key in (pygame.K_ESCAPE, pygame.K_q):
                    running = False
                elif event.key == pygame.K_r:
                    result = env.reset(domain_name="cartpole", task_name=task)
                    total_reward = 0.0
                    step_count = 0
                    print("Environment reset")

        # Check for held keys (for continuous control)
        keys = pygame.key.get_pressed()
        if keys[pygame.K_LEFT]:
            action_value = -1.0
        elif keys[pygame.K_RIGHT]:
            action_value = 1.0
        else:
            action_value = 0.0

        # Step the environment using OpenEnv pattern
        action = DMControlAction(values=[action_value])
        result = env.step(action)

        # Track reward from result
        total_reward += result.reward or 0.0
        step_count += 1

        # Check if episode is done
        if result.done:
            print(
                f"Episode finished! Steps: {step_count}, "
                f"Total reward: {total_reward:.2f}"
            )
            # Auto-reset on done
            result = env.reset(domain_name="cartpole", task_name=task)
            total_reward = 0.0
            step_count = 0

        # Update display
        direction = (
            "<--" if action_value < 0 else ("-->" if action_value > 0 else "---")
        )
        screen.fill((30, 30, 30))
        text = font.render(
            f"Step: {step_count} | Reward: {total_reward:.1f} | {direction}",
            True,
            (255, 255, 255),
        )
        screen.blit(text, (10, 40))
        pygame.display.flip()

        # Print progress periodically
        if step_count % 200 == 0 and step_count > 0:
            print(f"Step {step_count}: Total reward: {total_reward:.2f}")

        # Cap at 30 FPS
        clock.tick(30)

    pygame.quit()
    print(f"Session ended. Final reward: {total_reward:.2f}")


def run_visual(env: DMControlEnv, task: str = "balance"):
    """Run with pygame visualization showing rendered frames."""
    import base64
    import io

    import pygame

    print("\n=== Visual Mode (OpenEnv Step/Observation Pattern) ===")

    # Reset environment with rendering enabled
    result = env.reset(domain_name="cartpole", task_name=task, render=True)
    print(f"Initial observations: {list(result.observation.observations.keys())}")

    # Get first frame to determine window size
    if result.observation.pixels is None:
        print("Error: Server did not return rendered pixels.")
        print("Make sure the server supports render=True")
        print("\nTry running in interactive mode (default) instead.")
        sys.exit(1)

    # Decode base64 PNG to pygame surface
    png_data = base64.b64decode(result.observation.pixels)
    frame = pygame.image.load(io.BytesIO(png_data))
    frame_size = frame.get_size()

    # Initialize pygame
    pygame.init()
    screen = pygame.display.set_mode(frame_size)
    pygame.display.set_caption(
        "Cartpole (OpenEnv) - Arrow Keys to Move, R to Reset, ESC to Quit"
    )
    clock = pygame.time.Clock()

    print("Controls:")
    print("  LEFT/RIGHT arrows: Move cart")
    print("  R: Reset environment")
    print("  ESC or Q: Quit")

    running = True
    total_reward = 0.0
    step_count = 0

    while running:
        # Handle events
        for event in pygame.event.get():
            if event.type == pygame.QUIT:
                running = False
            elif event.type == pygame.KEYDOWN:
                if event.key in (pygame.K_ESCAPE, pygame.K_q):
                    running = False
                elif event.key == pygame.K_r:
                    result = env.reset(
                        domain_name="cartpole", task_name=task, render=True
                    )
                    total_reward = 0.0
                    step_count = 0
                    print("Environment reset")

        # Check for held keys (for continuous control)
        keys = pygame.key.get_pressed()
        if keys[pygame.K_LEFT]:
            action_value = -1.0
        elif keys[pygame.K_RIGHT]:
            action_value = 1.0
        else:
            action_value = 0.0

        # Step the environment using OpenEnv pattern
        action = DMControlAction(values=[action_value])
        result = env.step(action, render=True)

        # Track reward from result
        total_reward += result.reward or 0.0
        step_count += 1

        # Check if episode is done
        if result.done:
            print(
                f"Episode finished! Steps: {step_count}, "
                f"Total reward: {total_reward:.2f}"
            )
            result = env.reset(domain_name="cartpole", task_name=task, render=True)
            total_reward = 0.0
            step_count = 0

        # Render the frame from observation pixels
        if result.observation.pixels:
            png_data = base64.b64decode(result.observation.pixels)
            frame = pygame.image.load(io.BytesIO(png_data))
            screen.blit(frame, (0, 0))
            pygame.display.flip()

        # Print progress periodically
        if step_count % 200 == 0 and step_count > 0:
            print(f"Step {step_count}: Total reward: {total_reward:.2f}")

        # Cap at 30 FPS
        clock.tick(30)

    pygame.quit()
    print(f"Session ended. Final reward: {total_reward:.2f}")


def main():
    parser = argparse.ArgumentParser(
        description="Interactive cartpole control via OpenEnv"
    )
    parser.add_argument(
        "--visual",
        action="store_true",
        help="Enable pygame visualization with rendered frames",
    )
    parser.add_argument(
        "--headless",
        action="store_true",
        help="Run in headless mode (no pygame, automated control)",
    )
    parser.add_argument(
        "--max-steps",
        type=int,
        default=500,
        help="Maximum steps for headless mode (default: 500)",
    )
    parser.add_argument(
        "--task",
        type=str,
        default="balance",
        choices=["balance", "balance_sparse", "swingup", "swingup_sparse"],
        help="Cartpole task (default: balance)",
    )
    args = parser.parse_args()

    server_url = "http://localhost:8000"
    print(f"Connecting to {server_url}...")

    try:
        with DMControlEnv(base_url=server_url) as env:
            print("Connected!")

            # Get environment state
            state = env.state()
            print(f"Domain: {state.domain_name}, Task: {state.task_name}")
            print(f"Action spec: {state.action_spec}")

            if args.headless:
                run_headless(env, task=args.task, max_steps=args.max_steps)
            elif args.visual:
                run_visual(env, task=args.task)
            else:
                run_interactive(env, task=args.task)

    except ConnectionError as e:
        print(f"Failed to connect: {e}")
        print("\nMake sure the server is running:")
        print("  cd OpenEnv")
        print(
            "  PYTHONPATH=src:envs uvicorn envs.dm_control_env.server.app:app --port 8000"
        )
        sys.exit(1)


if __name__ == "__main__":
    main()