Reinforcement Learning
stable-baselines3
PyTorch
English
Korean
deep-reinforcement-learning
ppo
continuous-control
mujoco
pusher
pusher-v5
robotics
robot
robot-arm
robotic-manipulation
7-dof
gymnasium
Eval Results (legacy)
Instructions to use hwihwalab/pusher-v5-ppo with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- stable-baselines3
How to use hwihwalab/pusher-v5-ppo with stable-baselines3:
from huggingface_sb3 import load_from_hub checkpoint = load_from_hub( repo_id="hwihwalab/pusher-v5-ppo", filename="{MODEL FILENAME}.zip", ) - Notebooks
- Google Colab
- Kaggle
File size: 18,476 Bytes
f8364f0 | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 | """
app.py
Pusher-v5 ์ค์๊ฐ ์น ์๋ฎฌ๋ ์ด์
๋ฐ ํ์ต ๊ด์ FastAPI ๋ฐฑ์๋ ์๋ฒ
"""
import os
import io
import json
import time
import base64
import asyncio
import threading
from typing import Optional, Dict, Any
import numpy as np
from PIL import Image
import gymnasium as gym
from stable_baselines3 import PPO
import uvicorn
from fastapi import FastAPI, WebSocket, WebSocketDisconnect, HTTPException
from fastapi.responses import FileResponse, JSONResponse
from fastapi.staticfiles import StaticFiles
from fastapi.middleware.cors import CORSMiddleware
import visualizer
import train as train_module
app = FastAPI(title="Pusher-v5 Real-time Simulation & RL Control Center")
app.add_middleware(
CORSMiddleware,
allow_origins=["*"],
allow_credentials=True,
allow_methods=["*"],
allow_headers=["*"],
)
# Global simulation state
class SimulationManager:
def __init__(self):
self.env: Optional[gym.Env] = None
self.model: Optional[PPO] = None
self.model_path = "./results/ppo_pusher.zip"
self.is_running = False
self.is_paused = False
self.policy_type = "trained" # 'trained' or 'random'
self.show_hud = True
self.speed = 1.0
self.obs = None
self.ep_reward = 0.0
self.ep_step = 0
self.episode_count = 1
self.lock = threading.Lock()
self.clients = set()
# Load environment and model
self.init_env_and_model()
def init_env_and_model(self):
with self.lock:
try:
if self.env is not None:
self.env.close()
self.env = gym.make("Pusher-v5", render_mode="rgb_array")
self.obs, _ = self.env.reset(seed=42)
self.ep_reward = 0.0
self.ep_step = 0
if os.path.exists(self.model_path):
self.model = PPO.load(self.model_path, env=self.env)
print(f"[SimManager] Loaded trained model from {self.model_path}")
else:
self.model = None
print("[SimManager] Trained model not found, using random policy.")
except Exception as e:
print(f"[SimManager Init Error] {e}")
def reset_env(self, seed: Optional[int] = None):
with self.lock:
if self.env is None:
self.env = gym.make("Pusher-v5", render_mode="rgb_array")
seed_val = seed if seed is not None else int(time.time() * 1000) % 100000
self.obs, _ = self.env.reset(seed=seed_val)
self.ep_reward = 0.0
self.ep_step = 0
self.episode_count += 1
def step(self) -> Dict[str, Any]:
with self.lock:
if self.env is None or self.obs is None:
self.reset_env()
try:
if self.policy_type == "trained" and self.model is not None:
action, _ = self.model.predict(self.obs, deterministic=True)
else:
action = self.env.action_space.sample()
next_obs, reward, terminated, truncated, info = self.env.step(action)
self.ep_reward += float(reward)
self.ep_step += 1
done = terminated or truncated
reward_dist = float(info.get("reward_dist", 0.0))
reward_near = float(info.get("reward_near", 0.0))
reward_ctrl = float(info.get("reward_ctrl", 0.0))
dist_goal = abs(reward_dist) / 1.25 if reward_dist <= 0 else reward_dist
dist_arm = abs(reward_near) if reward_near <= 0 else reward_near
raw_frame = self.env.render()
if raw_frame is not None:
if self.show_hud:
frame = visualizer.draw_hud_on_frame(
frame=raw_frame,
timestep=self.ep_step,
episode=self.episode_count,
step=self.ep_step,
step_reward=float(reward),
total_reward=self.ep_reward,
reward_dist=reward_dist,
reward_near=reward_near,
actions=action,
tag=f"Live ({'PPO' if self.policy_type=='trained' else 'Random'})",
)
else:
frame = raw_frame
else:
frame = np.zeros((480, 480, 3), dtype=np.uint8)
# JPEG Encode
buf = io.BytesIO()
Image.fromarray(frame).save(buf, format="JPEG", quality=75)
frame_b64 = base64.b64encode(buf.getvalue()).decode("utf-8")
# Extract 3D telemetry
joint_angles = [float(x) for x in self.obs[0:7]] if len(self.obs) >= 7 else []
tip_pos = [float(x) for x in self.obs[14:17]] if len(self.obs) >= 17 else [0, 0, 0]
obj_pos = [float(x) for x in self.obs[17:20]] if len(self.obs) >= 20 else [0, 0, 0]
goal_pos = [float(x) for x in self.obs[20:23]] if len(self.obs) >= 23 else [0, 0, 0]
packet = {
"type": "sim_frame",
"frame": frame_b64,
"episode": self.episode_count,
"step": self.ep_step,
"step_reward": float(reward),
"ep_reward": self.ep_reward,
"dist_goal": dist_goal,
"dist_arm": dist_arm,
"reward_ctrl": reward_ctrl,
"joints": joint_angles,
"tip_pos": tip_pos,
"obj_pos": obj_pos,
"goal_pos": goal_pos,
"actions": [float(a) for a in action],
"policy": self.policy_type,
"done": done,
}
self.obs = next_obs
if done:
self.reset_env()
return packet
except Exception as e:
print(f"[SimManager Step Error] {e}")
self.reset_env()
return {"type": "sim_frame", "frame": "", "episode": self.episode_count, "step": 0, "step_reward": 0.0, "ep_reward": 0.0, "dist_goal": 0.0, "dist_arm": 0.0, "reward_ctrl": 0.0, "joints": [], "tip_pos": [0,0,0], "obj_pos": [0,0,0], "goal_pos": [0,0,0], "actions": [], "policy": self.policy_type, "done": True}
# Global Background Training Manager
class TrainingManager:
def __init__(self):
self.is_training = False
self.progress = 0
self.current_timesteps = 0
self.total_timesteps = 0
self.status = "Idle"
self.logs = []
self.thread = None
self.lock = threading.Lock()
def add_log(self, message: str):
timestamp = time.strftime("%I:%M:%S %p")
with self.lock:
self.logs.append(f"[{timestamp}] {message}")
if len(self.logs) > 200:
self.logs.pop(0)
def start_training(self, timesteps: int = 20000, eval_freq: int = 5000):
with self.lock:
if self.is_training:
return False, "Training already in progress."
self.is_training = True
self.progress = 0
self.current_timesteps = 0
self.total_timesteps = timesteps
self.status = "Training in progress..."
self.logs = []
self.add_log(f"[PPO Engine] Initializing 7-DOF MuJoCo environment...")
self.add_log(f"[PPO Setup] Target: {timesteps:,} Timesteps | Eval Frequency: Every {eval_freq:,} Steps")
def run():
try:
# Custom callback to log progress periodically
class LiveLogCallback(train_module.VisualProgressCallback):
def __init__(self, manager, *args, **kwargs):
super().__init__(*args, **kwargs)
self.manager = manager
self.start_t = time.time()
self.last_log_step = 0
def _on_step(self) -> bool:
res = super()._on_step()
step = self.num_timesteps
if step - self.last_log_step >= 1000:
self.last_log_step = step
elapsed = max(0.1, time.time() - self.start_t)
fps = int(step / elapsed)
pct = min(100.0, (step / max(1, self.manager.total_timesteps)) * 100)
with self.manager.lock:
self.manager.progress = int(pct)
self.manager.current_timesteps = step
self.manager.add_log(
f"[PPO Step {step:07d}] Progress: {pct:5.1f}% | Training Speed: {fps:4d} FPS"
)
return res
def _record_checkpoint(self, timestep: int, tag: str):
super()._record_checkpoint(timestep, tag)
last_r = self.metrics["eval_rewards"][-1] if self.metrics["eval_rewards"] else 0.0
last_d = self.metrics["eval_dist_goal"][-1] if self.metrics["eval_dist_goal"] else 0.0
self.manager.add_log(
f"[PPO Checkpoint] Step {timestep:07d} | Mean Return: {last_r:+.2f} | Goal Dist: {last_d:.3f}m | Rendered Video & GIF"
)
train_env = gym.make("Pusher-v5")
eval_env = gym.make("Pusher-v5", render_mode="rgb_array")
model = PPO(
policy="MlpPolicy",
env=train_env,
learning_rate=3e-4,
n_steps=2048,
batch_size=64,
n_epochs=10,
gamma=0.99,
verbose=0,
)
cb = LiveLogCallback(
manager=self,
eval_env=eval_env,
eval_freq=eval_freq,
output_dir="./results",
fps=30,
)
self.add_log("[PPO Train] Learning loop started...")
model.learn(total_timesteps=timesteps, callback=cb)
model.save("./results/ppo_pusher")
self.add_log("[Model Saved] Model weights saved to ./results/ppo_pusher.zip")
# Generate plots & HTML & ZIP
plot_path = "./results/plots/training_metrics.png"
visualizer.generate_training_plots(cb.metrics, plot_path)
dashboard_path = "./results/dashboard.html"
visualizer.generate_html_dashboard(
metrics=cb.metrics,
checkpoint_records=cb.records,
plot_rel_path="plots/training_metrics.png",
output_html_path=dashboard_path,
)
zip_out = os.path.abspath("./ppo_pusher_bundle.zip")
train_module.bundle_into_zip("./results", zip_out)
self.add_log(f"[Artifact Bundle] Packaged ppo_pusher_bundle.zip ({os.path.getsize(zip_out)/(1024*1024):.2f} MB)")
# Reload model in sim manager
sim_manager.init_env_and_model()
self.add_log("[PPO Ready] Simulation manager refreshed with latest trained weights.")
with self.lock:
self.status = "Completed successfully"
self.progress = 100
except Exception as e:
self.add_log(f"[PPO Error] {e}")
with self.lock:
self.status = f"Error: {e}"
finally:
with self.lock:
self.is_training = False
self.thread = threading.Thread(target=run, daemon=True)
self.thread.start()
return True, "Training started in background."
sim_manager = SimulationManager()
training_manager = TrainingManager()
# WebSocket endpoint for real-time simulation streaming
@app.websocket("/ws/simulation")
async def websocket_simulation(websocket: WebSocket):
await websocket.accept()
sim_manager.clients.add(websocket)
print(f"[WebSocket] Client connected: {websocket.client}")
try:
while True:
# Handle incoming commands if any (non-blocking)
try:
msg = await asyncio.wait_for(websocket.receive_text(), timeout=0.001)
data = json.loads(msg)
cmd = data.get("command")
if cmd == "start":
sim_manager.is_running = True
sim_manager.is_paused = False
elif cmd == "pause":
sim_manager.is_paused = not sim_manager.is_paused
elif cmd == "stop":
sim_manager.is_running = False
elif cmd == "step":
sim_manager.is_paused = True
packet = sim_manager.step()
await websocket.send_json(packet)
elif cmd == "reset":
sim_manager.reset_env()
packet = sim_manager.step()
await websocket.send_json(packet)
elif cmd == "set_policy":
sim_manager.policy_type = data.get("policy", "trained")
elif cmd == "toggle_hud":
sim_manager.show_hud = not sim_manager.show_hud
elif cmd == "set_speed":
sim_manager.speed = float(data.get("speed", 1.0))
except asyncio.TimeoutError:
pass
except json.JSONDecodeError:
pass
# If simulation is active and not paused, step and stream
if sim_manager.is_running and not sim_manager.is_paused:
packet = sim_manager.step()
await websocket.send_json(packet)
# Frame rate throttling: 30 FPS default
sleep_time = max(0.01, (1.0 / 30.0) / max(0.2, sim_manager.speed))
await asyncio.sleep(sleep_time)
else:
await asyncio.sleep(0.05)
except WebSocketDisconnect:
print(f"[WebSocket] Client disconnected: {websocket.client}")
except Exception as e:
print(f"[WebSocket Error] {e}")
finally:
sim_manager.clients.discard(websocket)
# REST Endpoints
@app.get("/api/metrics")
def get_metrics():
metrics_path = "./results/metrics.json"
if os.path.exists(metrics_path):
with open(metrics_path, "r", encoding="utf-8") as f:
return json.load(f)
return {"eval_timesteps": [], "eval_rewards": [], "eval_dist_goal": []}
@app.get("/api/checkpoints")
def get_checkpoints():
videos_dir = "./results/videos"
metrics_path = "./results/metrics.json"
metrics_map = {}
if os.path.exists(metrics_path):
try:
with open(metrics_path, "r", encoding="utf-8") as f:
m_data = json.load(f)
timesteps = m_data.get("eval_timesteps", [])
rewards = m_data.get("eval_rewards", [])
dists = m_data.get("eval_dist_goal", [])
for i, step in enumerate(timesteps):
metrics_map[step] = {
"reward": rewards[i] if i < len(rewards) else 0.0,
"dist_goal": dists[i] if i < len(dists) else 0.0,
}
except Exception:
pass
checkpoints = []
if os.path.exists(videos_dir):
files = sorted(os.listdir(videos_dir))
mp4s = [f for f in files if f.endswith(".mp4")]
for mp4 in mp4s:
base = mp4.replace(".mp4", "")
gif = f"{base}.gif" if f"{base}.gif" in files else None
try:
step_num = int(base.split("_")[-1])
except Exception:
step_num = 0
if metrics_map and step_num not in metrics_map:
continue
m_info = metrics_map.get(step_num, {})
reward_val = m_info.get("reward", None)
dist_val = m_info.get("dist_goal", None)
checkpoints.append({
"name": base,
"step": step_num,
"reward": reward_val,
"dist_goal": dist_val,
"mp4": f"/results/videos/{mp4}",
"gif": f"/results/videos/{gif}" if gif else None,
})
return {"checkpoints": checkpoints}
@app.get("/api/bundle/download")
def download_bundle():
zip_path = os.path.abspath("./ppo_pusher_bundle.zip")
if os.path.exists(zip_path):
return FileResponse(zip_path, filename="ppo_pusher_bundle.zip", media_type="application/zip")
raise HTTPException(status_code=404, detail="Bundle file not found. Train the model first.")
@app.post("/api/train/start")
def start_train(timesteps: int = 20000, eval_freq: int = 5000):
success, msg = training_manager.start_training(timesteps=timesteps, eval_freq=eval_freq)
return {"success": success, "message": msg}
@app.get("/api/train/status")
def get_train_status():
return {
"is_training": training_manager.is_training,
"progress": training_manager.progress,
"current_timesteps": training_manager.current_timesteps,
"total_timesteps": training_manager.total_timesteps,
"status": training_manager.status,
"logs": training_manager.logs,
}
# Mount Static and Results directories
os.makedirs("web", exist_ok=True)
os.makedirs("results", exist_ok=True)
os.makedirs("results/videos", exist_ok=True)
os.makedirs("results/plots", exist_ok=True)
app.mount("/results", StaticFiles(directory="results"), name="results")
app.mount("/", StaticFiles(directory="web", html=True), name="web")
def main():
print("=" * 65)
print(" Pusher-v5 Real-time Simulation & Control Center Web Server")
print(" -> Local Web URL: http://localhost:8000")
print("=" * 65)
uvicorn.run(app, host="0.0.0.0", port=8000)
if __name__ == "__main__":
main()
|