File size: 10,655 Bytes
50cd0bb | 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 | #!/usr/bin/env python
"""MuJoCo Tello simulator — real rigid-body physics + a real quadrotor controller.
Why MuJoCo and not Gazebo: the existing Tello Gazebo package (clydemcqueen/tello_ros) needs
ROS2 Foxy + Gazebo Classic, both EOL, and neither installs sanely on Apple Silicon. MuJoCo
ships native arm64 wheels (`pip install mujoco`) and needs no ROS. There is no official Tello
MJCF, so `assets/tello.xml` is written from the published spec (98×92.5×41 mm, 80 g, 3" props,
white shell + prop guards).
The drone is flown the way the real one is: you send high-level commands ("forward 30 cm") and
an onboard controller does the work. Here that controller is a standard cascade —
position PD -> desired acceleration -> desired tilt + thrust
-> attitude PD -> torques -> 4 rotor forces
so the model actually banks to translate and settles like a real quadrotor, instead of
teleporting like the kinematic `SimDrone`.
Exposes the SAME interface as SimDrone (connect/takeoff/land/hover/move/flip), so `SafePilot`
and the whole BCI stack work unchanged. Rendering is OFFSCREEN (returns RGB arrays), which
avoids macOS's `mjpython` requirement for the interactive viewer and lets us drop the FPV
image straight into the Qt stimulus window.
python sim_mujoco.py --demo # scripted flight, saves a contact sheet
python sim_mujoco.py --demo --live # + an OpenCV window if cv2 is available
"""
from __future__ import annotations
import argparse
import threading
import time
from pathlib import Path
import numpy as np
ASSETS = Path(__file__).resolve().parents[2] / "assets" / "tello.xml"
G = 9.81
class MuJoCoDrone:
"""Physics-backed Tello with the SimDrone command surface."""
def __init__(self, model_path=ASSETS, realtime=True, verbose=True):
import mujoco
self.mj = mujoco
self.model = mujoco.MjModel.from_xml_path(str(model_path))
self.data = mujoco.MjData(self.model)
self.mass = float(self.model.body_subtreemass[
mujoco.mj_name2id(self.model, mujoco.mjtObj.mjOBJ_BODY, "tello")])
self.bid = mujoco.mj_name2id(self.model, mujoco.mjtObj.mjOBJ_BODY, "tello")
self.realtime, self.verbose = realtime, verbose
self.flying = False
self.target = np.array([0.0, 0.0, 0.05]) # world-frame setpoint (m)
self.yaw_target = 0.0
self._lock = threading.Lock()
self._run = False
self._thread = None
self._renderers = {}
self.log = []
# control gains (tuned for an 80 g craft; gentle, BCI-friendly)
self.kp_pos, self.kd_pos = np.array([3.0, 3.0, 12.0]), np.array([2.6, 2.6, 5.0])
self.kp_att, self.kd_att = 0.06, 0.012
self.tilt_max = np.deg2rad(22)
# ------------------------------------------------------------------ state
@property
def pos(self):
return self.data.xpos[self.bid].copy()
@property
def vel(self):
return self.data.cvel[self.bid][3:].copy()
def rpy(self):
R = self.data.xmat[self.bid].reshape(3, 3)
return (np.arctan2(R[2, 1], R[2, 2]), # roll
-np.arcsin(np.clip(R[2, 0], -1, 1)), # pitch
np.arctan2(R[1, 0], R[0, 0])) # yaw
# ------------------------------------------------------------- controller
def _control(self):
"""Cascade: position PD -> tilt + thrust -> attitude PD -> rotor mix."""
p, v = self.pos, self.vel
roll, pitch, yaw = self.rpy()
if not self.flying:
self.data.ctrl[:] = 0.0
return
a_des = self.kp_pos * (self.target - p) - self.kd_pos * v # world accel
a_des[:2] = np.clip(a_des[:2], -6.0, 6.0)
# thrust magnitude: vertical component + gravity, projected on body z
T = self.mass * (G + a_des[2]) / max(np.cos(roll) * np.cos(pitch), 0.5)
T = float(np.clip(T, 0.0, 3.0))
# desired tilt (small-angle): +pitch tips thrust to +x, +roll tips it to -y
cy, sy = np.cos(self.yaw_target), np.sin(self.yaw_target)
ax, ay = a_des[0] * cy + a_des[1] * sy, -a_des[0] * sy + a_des[1] * cy
pitch_des = np.clip(ax / G, -self.tilt_max, self.tilt_max)
roll_des = np.clip(-ay / G, -self.tilt_max, self.tilt_max)
w = self.data.cvel[self.bid][:3] # angular velocity
tx = self.kp_att * (roll_des - roll) - self.kd_att * w[0]
ty = self.kp_att * (pitch_des - pitch) - self.kd_att * w[1]
yaw_err = (self.yaw_target - yaw + np.pi) % (2 * np.pi) - np.pi
tz = 0.02 * yaw_err - 0.006 * w[2]
L, k = 0.0455 * np.sqrt(2), 0.0075 # arm length (x-config), yaw reaction coeff
f = np.array([ # order matches actuators: FL, FR, BR, BL
T / 4 + tx / (4 * L) - ty / (4 * L) + tz / (4 * k),
T / 4 - tx / (4 * L) - ty / (4 * L) - tz / (4 * k),
T / 4 - tx / (4 * L) + ty / (4 * L) + tz / (4 * k),
T / 4 + tx / (4 * L) + ty / (4 * L) - tz / (4 * k)])
self.data.ctrl[:] = np.clip(f, 0.0, 0.8)
def step(self, n=1):
with self._lock:
for _ in range(n):
self._control()
self.mj.mj_step(self.model, self.data)
# --------------------------------------------------------------- commands
def _say(self, msg):
self.log.append((time.time(), msg))
if self.verbose:
p = self.pos
print(f" [mujoco] {msg:<20} pos=({p[0]:+.2f},{p[1]:+.2f},{p[2]:.2f})m")
def connect(self):
self._say("connect"); return True
def start(self, hz=500):
"""Run physics in a background thread (BCI loop stays responsive)."""
if self._thread:
return
self._run = True
def loop():
dt = self.model.opt.timestep
nxt = time.time()
while self._run:
self.step(1)
if self.realtime:
nxt += dt
s = nxt - time.time()
if s > 0:
time.sleep(s)
elif s < -0.5:
nxt = time.time()
self._thread = threading.Thread(target=loop, daemon=True); self._thread.start()
def stop(self):
self._run = False
if self._thread:
self._thread.join(timeout=1.0); self._thread = None
def takeoff(self):
self.flying = True
self.target = np.array([self.pos[0], self.pos[1], 0.8])
self._say("takeoff")
def land(self):
self.target = np.array([self.pos[0], self.pos[1], 0.02])
self._say("land")
t0 = time.time()
while self.pos[2] > 0.06 and time.time() - t0 < 4.0:
time.sleep(0.02)
self.flying = False
def hover(self):
self.target = self.pos.copy()
def move(self, direction, cm):
if not self.flying:
self._say(f"IGNORED {direction} (not flying)"); return
d = cm / 100.0
yaw = self.yaw_target
fwd = np.array([np.cos(yaw), np.sin(yaw), 0.0])
lft = np.array([-np.sin(yaw), np.cos(yaw), 0.0])
delta = {"forward": fwd * d, "back": -fwd * d, "left": lft * d, "right": -lft * d,
"up": np.array([0, 0, d]), "down": np.array([0, 0, -d])}[direction]
self.target = self.target + delta
self.target[2] = max(0.15, self.target[2])
self._say(f"{direction} {cm}cm")
def rotate(self, direction, deg):
if not self.flying:
self._say(f"IGNORED {direction} (not flying)"); return
d = np.deg2rad(deg)
self.yaw_target += -d if direction == "rotate_cw" else d # +yaw = CCW (right-hand)
self._say(f"{direction} {deg}deg -> yaw {np.rad2deg(self.yaw_target):+.0f}")
def flip(self, direction="f"):
self._say(f"flip {direction} (visual only in sim)")
def battery(self):
return 100
# ---------------------------------------------------------------- render
def render(self, camera="chase", width=480, height=360):
"""Offscreen RGB frame — no mjpython needed, drops straight into a Qt widget."""
key = (camera, width, height)
if key not in self._renderers:
self._renderers[key] = self.mj.Renderer(self.model, height=height, width=width)
r = self._renderers[key]
with self._lock:
r.update_scene(self.data, camera=camera)
return r.render()
# ----------------------------------------------------------------------- demo
def _demo(live=False, out="results/mujoco_flight.png"):
from drone import SafePilot, SafetyConfig # noqa: E402
d = MuJoCoDrone()
print(f"model loaded: mass={d.mass*1000:.0f} g, hover thrust={d.mass*G:.3f} N total\n")
d.connect(); d.start()
pilot = SafePilot(d, SafetyConfig(step_cm=40, min_interval=0.0, hover_after=1e9))
print(" BCI 起飞被拦截 ->", pilot.update("takeoff"))
d.takeoff(); time.sleep(2.0)
shots = [("takeoff", d.render("chase"))]
for cmd in ["forward", "forward", "left", "up", "right", "back"]:
pilot.update(cmd, 1.0)
time.sleep(1.6)
shots.append((cmd, d.render("chase")))
fpv = d.render("fpv")
d.land(); d.stop()
p = d.pos
print(f"\n 最终位置 x={p[0]:+.2f} y={p[1]:+.2f} z={p[2]:.2f} m")
import matplotlib; matplotlib.use("Agg")
import matplotlib.pyplot as plt
n = len(shots) + 1
fig, ax = plt.subplots(2, (n + 1) // 2, figsize=(3.1 * ((n + 1) // 2), 5.2))
fig.patch.set_facecolor("white")
for a, (title, img) in zip(ax.ravel(), shots + [("FPV camera", fpv)]):
a.imshow(img); a.set_title(title, fontsize=10); a.axis("off")
for a in ax.ravel()[n:]:
a.axis("off")
fig.suptitle("MuJoCo Tello — physics sim (chase cam + onboard FPV)", fontweight="bold")
fig.tight_layout()
outp = Path(__file__).resolve().parents[2] / out
outp.parent.mkdir(parents=True, exist_ok=True)
fig.savefig(outp, dpi=120, bbox_inches="tight", facecolor="white")
print("saved", outp)
def main():
ap = argparse.ArgumentParser(description=__doc__,
formatter_class=argparse.RawDescriptionHelpFormatter)
ap.add_argument("--demo", action="store_true")
ap.add_argument("--live", action="store_true", help="also show a live cv2 window")
args = ap.parse_args()
if args.demo:
_demo(args.live)
else:
ap.print_help()
if __name__ == "__main__":
main()
|