diff --git a/docs/conftest.py b/docs/conftest.py new file mode 100644 index 000000000..5d2b826d9 --- /dev/null +++ b/docs/conftest.py @@ -0,0 +1 @@ +collect_ignore_glob = ["gen_*.py"] diff --git a/docs/gen_drone_renders.py b/docs/gen_drone_renders.py new file mode 100644 index 000000000..d1bd3a8ff --- /dev/null +++ b/docs/gen_drone_renders.py @@ -0,0 +1,89 @@ +"""Render one image of every supported drone for the documentation. + +This script is executed by the mkdocs-gen-files plugin during ``mkdocs build`` or +``mkdocs serve``. The images are written to ``img/drones/.png`` in the virtual docs tree +and are never committed. Rendering is headless via EGL on Linux and CGL on macOS unless +``MUJOCO_GL`` is already set. +""" + +import os +import sys +from pathlib import Path + +if sys.platform == "linux": + os.environ.setdefault("MUJOCO_GL", "egl") + # libglvnd only searches /usr/share/glvnd, which does not exist on a minimal CI runner. Append + # the Mesa driver of this environment so EGL still finds a (software) device there. + _vendors = Path(sys.prefix, "share/glvnd/egl_vendor.d") + if _vendors.is_dir(): + os.environ.setdefault( + "__EGL_VENDOR_LIBRARY_DIRS", f"/usr/share/glvnd/egl_vendor.d:{_vendors}" + ) +elif sys.platform == "darwin": + os.environ.setdefault("MUJOCO_GL", "cgl") + +import imageio.v3 as iio +import mkdocs_gen_files +import mujoco +import numpy as np + +from crazyflow.drones import Drone + +DRONE_DIR = Path("crazyflow/drones") +WIDTH, HEIGHT = 2000, 1500 +MARGIN = 80 + + +def render(name: str) -> np.ndarray: + """Render a drone in front of a transparent background, cropped to its bounding box.""" + spec = mujoco.MjSpec.from_file(str(DRONE_DIR / f"{name}.xml")) + for body in list(spec.worldbody.bodies): + if body.name != "drone": + spec.delete(body) + for material in spec.materials: + if 0 < material.rgba[3] < 1: + material.rgba[3] = 1 + spec.visual.global_.offwidth = WIDTH + spec.visual.global_.offheight = HEIGHT + spec.visual.quality.shadowsize = 8192 + spec.visual.quality.offsamples = 16 + spec.visual.headlight.ambient = [0.35, 0.35, 0.35] + spec.visual.headlight.diffuse = [0.5, 0.5, 0.5] + spec.worldbody.add_light( + pos=[0.4, -0.3, 0.6], dir=[-0.55, 0.4, -0.8], diffuse=[0.6, 0.6, 0.6], castshadow=True + ) + spec.worldbody.add_light( + pos=[-0.4, 0.3, 0.3], dir=[0.6, -0.45, -0.45], diffuse=[0.3, 0.3, 0.3], castshadow=False + ) + model = spec.compile() + data = mujoco.MjData(model) + mujoco.mj_forward(model, data) + + camera = mujoco.MjvCamera() + camera.lookat[:] = model.body("drone").pos + camera.distance = max(0.28, 1.1 * model.stat.extent) + camera.azimuth, camera.elevation = 135, -28 + option = mujoco.MjvOption() + option.geomgroup[:] = 0 + option.geomgroup[2] = 1 + + renderer = mujoco.Renderer(model, height=HEIGHT, width=WIDTH) + try: + renderer.update_scene(data, camera, option) + rgb = renderer.render() + renderer.enable_segmentation_rendering() + renderer.update_scene(data, camera, option) + mask = renderer.render()[..., 0] != -1 + finally: + renderer.close() + + rows, cols = np.nonzero(mask) + y0, y1 = max(rows.min() - MARGIN, 0), min(rows.max() + MARGIN, HEIGHT) + x0, x1 = max(cols.min() - MARGIN, 0), min(cols.max() + MARGIN, WIDTH) + alpha = mask.astype(np.uint8) * 255 + return np.dstack([rgb, alpha])[y0:y1, x0:x1] + + +for drone in Drone: + with mkdocs_gen_files.open(f"img/drones/{drone}.png", "wb") as fd: + fd.write(iio.imwrite("", render(drone), extension=".png")) diff --git a/docs/get-started/installation.md b/docs/get-started/installation.md index 9fe2d45ad..df5c9a757 100644 --- a/docs/get-started/installation.md +++ b/docs/get-started/installation.md @@ -75,7 +75,7 @@ Adds `pytest` and `pytest-markdown-docs` for running the test suite and doc snip ```bash uv run pytest -v tests # unit and integration tests - uv run pytest -v --markdown-docs --markdown-docs-syntax=superfences crazyflow/ docs/ --ignore=docs/gen_ref_pages.py + uv run pytest -v --markdown-docs --markdown-docs-syntax=superfences crazyflow/ docs/ ``` ## Verify the installation diff --git a/docs/index.md b/docs/index.md index bc2b5b1c5..c7587aa5d 100644 --- a/docs/index.md +++ b/docs/index.md @@ -35,6 +35,16 @@ Crazyflow is a research simulator for Crazyflie-style quadrotors that runs milli >
Obstacle avoidance with MPPI using Crazyflow for rollouts.
+
+ +
End-to-end vision-based flight on hardware with a policy trained in Crazyflow.
+