Spaces:
Runtime error
Runtime error
File size: 10,099 Bytes
6083286 |
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 |
"""
A 2D drone environment with dynamic wind forces for reinforcement learning.
The drone can apply discrete thrust actions while being affected by smoothly varying wind.
The goal is to navigate and survive within the bounded world.
"""
import numpy as np
import gymnasium as gym
from gymnasium import spaces
from typing import Optional, Tuple, Dict, Any
# Constants
DT = 0.1 # Time step
MAX_VEL = 2.0 # Maximum velocity magnitude
WIND_MAX = 2.0 # Maximum wind magnitude
WIND_SMOOTHING = 0.05 # Wind interpolation rate toward target
WIND_TARGET_INTERVAL = 50 # Steps between sampling new wind target
MAX_STEPS = 500 # Maximum episode length
POSITION_MIN = 0.0 # Minimum position (x, y)
POSITION_MAX = 1.0 # Maximum position (x, y)
THRUST = 0.25 # Thrust magnitude per action (slightly higher for control authority)
# Target zone (box) constants
TARGET_X_MIN = 0.7 # Target box left edge
TARGET_X_MAX = 0.9 # Target box right edge
TARGET_Y_MIN = 0.3 # Target box bottom edge
TARGET_Y_MAX = 0.7 # Target box top edge
TARGET_REWARD = 2.0 # Bonus reward for being in target zone
TARGET_SPAWN_DELAY = 50 # Steps before target zone appears (after wind starts)
# Stabilization and shaping
DRAG_COEFF = 0.3 # Linear velocity drag coefficient
SPEED_PENALTY_COEFF = 0.05 # Penalize high speeds to encourage smooth control
EDGE_MARGIN = 0.06 # Margin near boundaries where penalty increases
EDGE_PENALTY_COEFF = 0.5 # Strength of boundary proximity penalty
class DroneWindEnv(gym.Env):
"""
A 2D drone environment with dynamic wind.
Observation: [x, y, vx, vy, wind_x, wind_y]
Action: Discrete(5) - 0: no thrust, 1: up, 2: down, 3: left, 4: right
"""
metadata = {"render_modes": ["human", "rgb_array"], "render_fps": 10}
def __init__(self):
super().__init__()
# Observation space: [x, y, vx, vy, wind_x, wind_y]
self.observation_space = spaces.Box(
low=np.array([POSITION_MIN, POSITION_MIN, -MAX_VEL, -MAX_VEL, -WIND_MAX, -WIND_MAX], dtype=np.float32),
high=np.array([POSITION_MAX, POSITION_MAX, MAX_VEL, MAX_VEL, WIND_MAX, WIND_MAX], dtype=np.float32),
dtype=np.float32
)
# Action space: 5 discrete thrust directions
self.action_space = spaces.Discrete(5)
# Internal state
self.x: float = 0.0
self.y: float = 0.0
self.vx: float = 0.0
self.vy: float = 0.0
self.wind_x: float = 0.0
self.wind_y: float = 0.0
self.wind_target_x: float = 0.0
self.wind_target_y: float = 0.0
self.step_count: int = 0
def reset(
self,
*,
seed: Optional[int] = None,
options: Optional[Dict[str, Any]] = None
) -> Tuple[np.ndarray, Dict[str, Any]]:
"""
Reset the environment to initial state.
Args:
seed: Optional random seed
options: Optional reset options
Returns:
observation: Initial observation array
info: Empty info dict
"""
# Always call super().reset to ensure seeding and np_random are initialized
super().reset(seed=seed)
# Initialize state
self.x = 0.1
self.y = 0.5
self.vx = 0.0
self.vy = 0.0
self.wind_x = 0.0
self.wind_y = 0.0
self.wind_target_x = 0.0
self.wind_target_y = 0.0
self.step_count = 0
# Build observation
obs = self._get_observation()
info = {}
return obs, info
def step(self, action: int) -> Tuple[np.ndarray, float, bool, bool, Dict[str, Any]]:
"""
Execute one environment step.
Args:
action: Discrete action (0-4)
Returns:
observation: New observation array
reward: Reward for this step
terminated: Whether episode ended due to boundary crash
truncated: Whether episode ended due to max steps
info: Info dict with step_count
"""
# Increment step count
self.step_count += 1
# Update wind model
self._update_wind()
# Apply physics update
self._apply_physics(action)
# Compute reward
base_reward = 1.0 # Survival reward
# Check if drone is in target zone (only if target has spawned)
target_spawned = self.step_count >= TARGET_SPAWN_DELAY
in_target = False
if target_spawned:
in_target = (
TARGET_X_MIN <= self.x <= TARGET_X_MAX and
TARGET_Y_MIN <= self.y <= TARGET_Y_MAX
)
target_bonus = TARGET_REWARD if in_target else 0.0
# Speed penalty (discourage excessive velocity)
speed_sq = self.vx * self.vx + self.vy * self.vy
speed_penalty = -SPEED_PENALTY_COEFF * float(speed_sq)
# Boundary proximity penalty (discourage hovering near walls)
dist_left = self.x - POSITION_MIN
dist_right = POSITION_MAX - self.x
dist_bottom = self.y - POSITION_MIN
dist_top = POSITION_MAX - self.y
min_dist = min(dist_left, dist_right, dist_bottom, dist_top)
edge_penalty = 0.0
if min_dist < EDGE_MARGIN:
edge_penalty = -EDGE_PENALTY_COEFF * (EDGE_MARGIN - float(min_dist)) / EDGE_MARGIN
reward = base_reward + target_bonus + speed_penalty + edge_penalty
# Check termination (boundary crash)
terminated = (
self.x <= POSITION_MIN or
self.x >= POSITION_MAX or
self.y <= POSITION_MIN or
self.y >= POSITION_MAX
)
# Check truncation (max steps)
truncated = self.step_count >= MAX_STEPS
# Build observation
obs = self._get_observation()
# Check if in target zone (only if target has spawned)
target_spawned = self.step_count >= TARGET_SPAWN_DELAY
in_target = False
if target_spawned:
in_target = (
TARGET_X_MIN <= self.x <= TARGET_X_MAX and
TARGET_Y_MIN <= self.y <= TARGET_Y_MAX
)
info = {"step_count": self.step_count, "in_target": in_target, "target_spawned": target_spawned}
return obs, reward, terminated, truncated, info
def _update_wind(self) -> None:
"""Update wind by smoothly moving toward target, resampling target periodically."""
# Resample wind target every WIND_TARGET_INTERVAL steps
if self.step_count % WIND_TARGET_INTERVAL == 0:
self.wind_target_x = self.np_random.uniform(-WIND_MAX, WIND_MAX)
self.wind_target_y = self.np_random.uniform(-WIND_MAX, WIND_MAX)
# Smoothly interpolate wind toward target
self.wind_x += WIND_SMOOTHING * (self.wind_target_x - self.wind_x)
self.wind_y += WIND_SMOOTHING * (self.wind_target_y - self.wind_y)
# Clamp wind to bounds
self.wind_x = np.clip(self.wind_x, -WIND_MAX, WIND_MAX)
self.wind_y = np.clip(self.wind_y, -WIND_MAX, WIND_MAX)
def _apply_physics(self, action: int) -> None:
"""Apply physics update: convert action to thrust, update velocity and position."""
# Convert action to thrust vector
if action == 0: # No thrust
ax, ay = 0.0, 0.0
elif action == 1: # Thrust up
ax, ay = 0.0, THRUST
elif action == 2: # Thrust down
ax, ay = 0.0, -THRUST
elif action == 3: # Thrust left
ax, ay = -THRUST, 0.0
elif action == 4: # Thrust right
ax, ay = THRUST, 0.0
else:
raise ValueError(f"Invalid action: {action}. Must be in [0, 4]")
# Update velocity with thrust and wind
self.vx = self.vx + ax + self.wind_x * DT
self.vy = self.vy + ay + self.wind_y * DT
# Apply linear drag (proportional to velocity) for stability
self.vx -= DRAG_COEFF * self.vx * DT
self.vy -= DRAG_COEFF * self.vy * DT
# Clamp velocity
self.vx = np.clip(self.vx, -MAX_VEL, MAX_VEL)
self.vy = np.clip(self.vy, -MAX_VEL, MAX_VEL)
# Update position
self.x = self.x + self.vx * DT
self.y = self.y + self.vy * DT
# Clamp position to bounds
self.x = np.clip(self.x, POSITION_MIN, POSITION_MAX)
self.y = np.clip(self.y, POSITION_MIN, POSITION_MAX)
def _get_observation(self) -> np.ndarray:
"""Build observation array from current state."""
return np.array(
[self.x, self.y, self.vx, self.vy, self.wind_x, self.wind_y],
dtype=np.float32
)
def render(self) -> None:
"""
Render the environment state (stub implementation for Phase 1).
Prints state to stdout.
"""
print(
f"Step {self.step_count}: "
f"x={self.x:.2f}, y={self.y:.2f}, "
f"vx={self.vx:.2f}, vy={self.vy:.2f}, "
f"wind=({self.wind_x:.2f}, {self.wind_y:.2f})"
)
def make_drone_env() -> DroneWindEnv:
"""Helper function to create a DroneWindEnv instance."""
return DroneWindEnv()
if __name__ == "__main__":
# Manual test block
print("Testing DroneWindEnv...")
print("=" * 60)
env = make_drone_env()
obs, info = env.reset(seed=42)
print(f"Initial observation: {obs}")
print()
for t in range(200):
action = env.action_space.sample()
obs, reward, terminated, truncated, info = env.step(action)
env.render()
if terminated:
print(f"\nEpisode terminated at step {t} (boundary crash)")
break
if truncated:
print(f"\nEpisode truncated at step {t} (max steps reached)")
break
print("=" * 60)
print("Test completed!")
|