File size: 10,966 Bytes
80c0c44 77e15e7 d0e61c2 77e15e7 80c0c44 d0e61c2 80c0c44 33523d4 d0e61c2 33523d4 d0e61c2 33523d4 80c0c44 0863c0a 80c0c44 0863c0a 80c0c44 0863c0a 80c0c44 | 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 | """
server/grid_world_environment.py
Core Drone Delivery RL Environment — refactored to use core/ modules.
"""
from __future__ import annotations
import random
from typing import Any, Dict, List, Optional, Tuple
import torch
try:
from openenv.core.env_server import Environment
except ImportError:
class Environment:
def reset(self): raise NotImplementedError
def step(self, action): raise NotImplementedError
@property
def state(self): raise NotImplementedError
from drone_env.models import DroneAction, DroneObservation, DroneState
from drone_env.core.grid_generator import generate_city_map, EMOJI, LEGEND
from drone_env.core.tasks import TASK_CONFIG
from drone_env.core.drone import compute_next_pos, drain_battery
from drone_env.core.obstacles import check_move
from drone_env.core.state_manager import new_episode_state
try:
from drone_env.graders import GRADERS
except ImportError:
try:
from graders import GRADERS
except ImportError:
# Final fallback
GRADERS = {}
from drone_env.rl.trainer import record_episode
class DroneDeliveryEnvironment(Environment):
def __init__(self):
super().__init__()
self._state = DroneState()
self._cfg = TASK_CONFIG["graders:grade_easy"]
self._grid: List[List[str]] = []
self._deliveries: List[Tuple[int, int]] = []
self._delivered: List[bool] = []
self._drone_x = 0
self._drone_y = 0
self._battery = 0
self._rng = torch.Generator()
self._step_records: List[Dict] = []
self._start_pos: Tuple[int, int] = (0, 0)
def reset(self, action: Optional[DroneAction] = None) -> DroneObservation:
# Potentially persist old episode if not already done
if self._step_records and self._state.episode_id and not self._state.done:
self._persist_episode()
# Map short task IDs (from openenv.yaml) to full grader keys
_TASK_ID_MAP = {
"easy_delivery": "graders:grade_easy",
"medium_delivery": "graders:grade_medium",
"hard_delivery": "graders:grade_hard",
# Legacy redirects
"drone_env.graders.easy:grade_easy": "graders:grade_easy",
"drone_env.graders.medium:grade_medium": "graders:grade_medium",
"drone_env.graders.hard:grade_hard": "graders:grade_hard",
}
task = "graders:grade_easy"
if action and action.task_name:
name = action.task_name
if name in _TASK_ID_MAP:
task = _TASK_ID_MAP[name]
elif name in TASK_CONFIG:
task = name
else:
# handle partial matches like "easy"
for k in TASK_CONFIG:
if k.startswith(name):
task = k; break
self._cfg = TASK_CONFIG[task]
self._rng.manual_seed(random.randint(0, 2**31))
self._grid, self._deliveries, self._start_pos = generate_city_map(self._cfg, self._rng)
self._delivered = [False] * len(self._deliveries)
self._drone_x, self._drone_y = self._start_pos
self._battery = self._cfg["battery_max"]
self._step_records = []
self._state = new_episode_state(
task=task,
deliveries=self._deliveries,
start_x=self._drone_x,
start_y=self._drone_y,
battery_max=self._cfg["battery_max"],
)
return self._build_obs(0.0, "🚁 Mission started! Deliver all packages.")
def step(self, action: DroneAction) -> DroneObservation:
if self._state.done:
return self._build_obs(0.0, "Episode ended. Call reset().")
cfg = self._cfg
direction = (action.direction or "WAIT").upper()
nx, ny = compute_next_pos(self._drone_x, self._drone_y, direction)
outcome, cell = check_move(self._grid, nx, ny, cfg["width"], cfg["height"])
reward = cfg["r_step"]
msg = ""
# Reward shaping (Distance bonus)
_, dist_before = self._next_target()
if outcome == "wall":
nx, ny = self._drone_x, self._drone_y
reward += cfg["r_wall"]
msg = f"Hit {direction} boundary! 🚫"
elif outcome == "blocked":
nx, ny = self._drone_x, self._drone_y
reward += cfg["r_blocked"]
msg = f"Path is {direction}ly blocked!"
elif outcome == "building":
self._drone_x, self._drone_y = nx, ny
reward = cfg.get("r_building", -0.1)
msg = f"Flying above a building! {EMOJI['building']}"
elif outcome == "tree":
self._drone_x, self._drone_y = nx, ny
reward = cfg.get("r_tree", -0.1)
msg = f"Flying over a tree! {EMOJI['tree']}"
elif outcome == "obstacle":
self._drone_x, self._drone_y = nx, ny
reward += cfg["r_obstacle"]
msg = f"Hit obstacle! {EMOJI['obstacle']}"
else:
self._drone_x, self._drone_y = nx, ny
if direction == "WAIT":
reward = cfg.get("r_wait", -0.1)
msg = "Drone is idling (WAIT)... 🔋"
elif not msg:
# Get cell type at current position
current_cell = self._grid[self._drone_y][self._drone_x]
if current_cell == "road":
msg = f"On road {EMOJI['road']}"
else:
msg = f"Flying over {current_cell} {EMOJI.get(current_cell, '')}"
# Move finalized, now check new distance
_, dist_after = self._next_target()
if dist_after is not None and dist_before is not None and outcome not in ["tree", "building", "ok"]:
# Reward for moving closer, small penalty for moving away
delta = dist_before - dist_after
reward += delta * 0.05
# Delivery check
for i, (tx, ty) in enumerate(self._deliveries):
if not self._delivered[i] and self._drone_x == tx and self._drone_y == ty:
self._delivered[i] = True
reward += cfg["r_delivery"]
self._state.deliveries_done += 1
msg = f"✅ Delivery {self._state.deliveries_done}/{len(self._deliveries)} done!"
break
# Battery drain
self._battery = drain_battery(self._battery, cfg["battery_cost"])
self._state.step_count += 1
self._state.reward_total += reward
bat_norm = max(0.0, float(self._battery) / cfg["battery_max"])
self._state.battery = bat_norm
self._state.drone_x = self._drone_x
self._state.drone_y = self._drone_y
self._step_records.append(dict(
step=self._state.step_count,
x=self._drone_x, y=self._drone_y,
action=direction,
reward=float(round(reward, 5)),
battery=float(round(bat_norm, 4)),
message=msg,
))
self._state.path_history = self._step_records
# Done conditions
done = False
if all(self._delivered):
done = True; msg = "🎉 All deliveries complete!"
elif self._battery <= 0:
reward += cfg["r_battery_dead"]
self._state.reward_total += cfg["r_battery_dead"]
done = True; msg = "🔋 Battery dead!"
elif self._state.step_count >= cfg["max_steps"]:
done = True; msg = "⏰ Max steps reached."
self._state.done = done
if done:
self._persist_episode()
return self._build_obs(reward, msg)
@property
def state(self) -> DroneState:
return self._state
@property
def graders(self) -> Dict:
"""Expose graders for the environment."""
return GRADERS
def _persist_episode(self):
try:
cfg = self._cfg
record_episode(
task=self._state.task_name,
steps=list(self._step_records),
grid_meta=dict(
width=cfg["width"], height=cfg["height"]
),
delivery_positions=[[d[0], d[1]] for d in self._deliveries],
deliveries_done=self._state.deliveries_done,
total_reward=float(round(self._state.reward_total, 4)),
)
except Exception as e:
print(f"[DroneEnv] Episode record error: {e}")
def _build_obs(self, reward: float, message: str) -> DroneObservation:
cfg = self._cfg
target, dist = self._next_target()
return DroneObservation(
grid=self._render_grid(),
grid_width=int(cfg["width"]), grid_height=int(cfg["height"]),
drone_x=self._drone_x, drone_y=self._drone_y,
battery=float(max(0.0, float(self._battery) / cfg["battery_max"])),
battery_steps_remaining=int(self._battery),
deliveries_total=int(len(self._deliveries)),
deliveries_done=int(self._state.deliveries_done),
current_target=target,
distance_to_target=float(dist) if dist is not None else None,
step_count=int(self._state.step_count),
max_steps=int(cfg["max_steps"]),
reward_last=float(max(0.01, min(0.99, round(reward, 4)))),
reward_total=float(max(0.01, min(0.99, round(self._state.reward_total, 4)))),
score=float(max(0.01, min(0.99, round(GRADERS[self._state.task_name](self._state), 4)))),
done=bool(self._state.done),
message=str(message),
legend=dict(LEGEND),
cell_types=list(self._grid),
)
def _render_grid(self) -> List[str]:
rows = []
for y, row in enumerate(self._grid):
line = ""
for x, cell in enumerate(row):
if x == self._drone_x and y == self._drone_y:
line += EMOJI["drone"]
continue
# Check for delivery targets
is_delivery = False
for i, (tx, ty) in enumerate(self._deliveries):
if x == tx and y == ty:
line += EMOJI["done_del"] if self._delivered[i] else EMOJI["delivery"]
is_delivery = True
break
if not is_delivery:
# Map the internal symbol to emoji
# Now it looks up from EMOJI dict using the symbolic string in 'cell'
line += EMOJI.get(cell, cell)
rows.append(line)
return rows
def _next_target(self):
for i, (tx, ty) in enumerate(self._deliveries):
if not self._delivered[i]:
dist = float(abs(self._drone_x - tx) + abs(self._drone_y - ty))
return (int(tx), int(ty)), float(dist)
return None, None
|