hyperdrone

Drone simulation stack for RLtools: raytracing renderer, L2F multirotor dynamics, and environment setup under one Python package.

Subpackage	What it is
hyperdrone.render	Raytracing renderer (OptiX / Metal / Vulkan / generic CPU). Self-contained — no dependency on drone dynamics.
hyperdrone.dynamics	Vectorized, stateful L2F multirotor simulator (Sim); cpu and cuda variants.
hyperdrone.env	The RL environment: the C++ MultiEnvironment<hyperdrone::World> driven through the exact rl_tools batch verbs.
hyperdrone.gym	Optional Gymnasium VectorEnv adapter over hyperdrone.env (pip install "hyperdrone[gym]").
hyperdrone.jit	Shared compile-and-cache infrastructure both domain packages build on.
hyperdrone.cuda	CUDA staging helpers (upload → DLPack tensor sets).

Native components are JIT-compiled per set of compile-time constants (resolution, camera count, drone count, ...) into a persistent cache; later uses load the cached library directly. render and dynamics are peers that never import each other — the coupling surface is a plain DLPack tensor of packed camera bases, device-resident on OptiX + CUDA. Rendering uses a shared scene module and a directly bound Python extension per renderer configuration; scene and asset-pool objects can be reused across configurations. Environment configurations are also directly bound extensions, sharing their C++ implementation with the golden-rollout generator.

Notebooks

Notebook	Open in Colab
HyperDrone
Benchmark
Custom
Orbiter

Install

From an rl-tools checkout (development):

.venv/bin/pip install -e interface/python/conta -e interface/python/hyperdrone

From an sdist (self-contained — bundles the rl-tools headers and raytracing sources, no checkout needed):

pip install hyperdrone-<version>.tar.gz

Install the dependencies used by the packaged examples with:

pip install "hyperdrone[examples]"

Requirements: CMake >= 3.24, a C++17 compiler, and git for dependency downloads. GLB loading uses the Assimp 6.0.4 release archive with a pinned SHA-256, shared by render and env builds. Cached archives are reused offline and verified by CMake before extraction. CUDA + OptiX driver is required for the OptiX render backend and cuda dynamics, Vulkan dev + glslang for the VULKAN backend. The WEBGPU backend fetches a hash-pinned wgpu-native prebuilt at first build (no extra system packages). Nothing beyond the build toolchain for GENERIC + cpu.

Environment

Variable	Meaning
HYPERDRONE_RENDER_BACKEND	OPTIX | METAL | VULKAN | WEBGPU | GENERIC | AUTO (default: Metal on macOS, OptiX elsewhere)
HYPERDRONE_DYNAMICS_DEVICE	CPU | CUDA | AUTO (default: cuda when available)
HYPERDRONE_CACHE_DIR	root for CMake build trees and downloaded build dependencies (default ~/.cache/hyperdrone)
HYPERDRONE_RLTOOLS_ROOT	rl-tools source root override (default: enclosing checkout, else the vendored tree)
HYPERDRONE_PROCTHOR_PATH	local ProcTHOR example scene override (otherwise downloaded and cached)
HYPERDRONE_SKIP_BUILD	skip the CMake staleness check when the artifacts already exist
HYPERDRONE_OFFLINE	forbid network during builds (requires seeded/vendored dependencies)
HYPERDRONE_BUILD_JOBS	parallel build jobs (default 5)

Each component and variant gets its own build tree (render-optix-*, dynamics-cuda-*, ...); switching never invalidates another's cache. All configure/build steps run under a per-tree file lock, so many worker processes can share one cache safely. CMake FetchContent sources and native build state live in <cache>/.dependencies; the native build does not write generated files into an editable checkout or installed package.

Whenever a renderer is successfully allocated, RLtools reports the selected backend on stderr, for example #rl_tools::rendering::raytracing: backend=metal. The Renderer.backend property provides the same lowercase name programmatically.

Rendering

import numpy as np
from hyperdrone import render
from hyperdrone.examples.data import procthor_scene_path

scene = render.load_scene(procthor_scene_path(), fidelity="high")

renderer = render.Renderer(width=320, height=240, num_cameras=4, output="rgbd", fidelity="high")
renderer.init(scene)

camera = renderer.camera(position=(0, 0, 1.5), look_at=(1, 0, 1.5), fov=80)
renderer.set_cameras(np.repeat(camera[None], 4, axis=0))
renderer.render()

rgb   = renderer.frame()    # (4, 240, 320, 4) uint8
depth = renderer.depth()    # (4, 240, 320) float32

The renderer operates in the L2F FLU frame: +X forward, +Y left, +Z up; GLB assets are swizzled at load time. Procedural scenes (render.Scene() + render.Object / render.Mesh / render.SceneLight), segmentation output, overlays (per-camera dynamic content via AssetPool + spawn/attach), motion blur, and anti-aliasing follow the same API as before under hyperdrone.render.*.

Converted ProcTHOR scenes carry ambient (0.6, 0.6, 0.6) and a black background in scenes[].extras.rl_tools.environment (mode: "solid", ambient and background are nonnegative linear RGB triples). Scene loading imports these settings; importing objects or assemblies leaves the parent environment alone. Ambient affects high and veryhigh, approximating indirect illumination without light bounces. Override imported settings with scene.set_environment(...) before renderer.init(scene).

render() produces all image outputs selected at construction; render_launch() and render_sync() split submission from waiting. Collision probes run separately through probe() or probe_launch() / probe_sync(), after generate_probe_directions(), and are read with collisions(). num_probes sets their count per camera (default 1); zero compiles the probe pass out and makes probe operations raise.

Zero-copy I/O (DLPack)

renderer.frame()                  # snapshot copy (safe to keep)
renderer.frame(copy=False)        # numpy view of the staging buffer (refreshed in place)
rgba = torch.from_dlpack(renderer.frame_dlpack())
# live uint8 (num_cameras, height, width, 4), zero-copy GPU tensor on OptiX
packed = torch.from_dlpack(renderer.frame_raw_dlpack())
# same memory as packed uint32 (num_cameras, height, width)

Inputs accept any DLPack producer. CUDA-resident camera input (OptiX): set_cameras dispatches on __dlpack_device__, so a CUDA tensor — a torch GPU tensor, a set pre-uploaded with hyperdrone.cuda.upload, or Sim.camera_bases() — is handed over device-to-device on the render stream, fully async; stream= takes the producer's cudaStream_t handle for event-ordered hand-off.

Dynamics

from hyperdrone import dynamics

sim = dynamics.Sim(num_drones=4096, model="x500_sim", device="cuda")
sim.reset(seed=0)                     # deterministic; host-sampled, identical on cpu/cuda
sim.step(actions)                     # (N, 4) float32 in [-1, 1], host or CUDA tensor
sim.state["position"]                 # zero-copy DLPack views, CUDA-resident on cuda
sim.state.numpy("position")           # host copy
sim.observe()                         # (N, 18): position, rotation matrix, velocities
sim.parameters["mass"] = masses       # per-drone runtime parameters (until the next reset)
sim.model_parameters()                # the preset's nominal l2f dynamics: rotor layout, mass, J, ...

cameras = sim.camera_bases(fov=100, aspect=renderer.aspect)
renderer.set_cameras(cameras, stream=sim.stream)   # device-resident hand-off

num_drones and domain randomization are compile-time (the JIT key); the model preset (crazyflie, x500_real, x500_sim, mrs, ...), integration dt, and all physical parameters are runtime. Reward and termination live in the environment (hyperdrone.env), not the simulator. One process can use one dynamics variant (cpu or cuda).

RL environment

import numpy as np
from hyperdrone.env import EnvConfig, MultiEnvironment

config = EnvConfig(num_environments=1, instances=1024, cam_width=32, cam_height=32, task="target_frame")
env = MultiEnvironment("scenes/", config=config, seed=0)

mask = np.ones(env.total_instances, dtype=bool)
env.reset(mask)                    # resample parameters + states where mask is set
env.render(mask)                   # render the FPV cameras (mask marks fresh episodes)
observations = env.observe()       # (total, observation_dim) float32
env.step(actions)                  # (total, action_dim) float32 in [-1, 1]
rewards, terminated = env.rewards(), env.terminated()
critic_input = env.observe_privileged()
env.rotate_scene()                 # deterministic scene rotation; reset all instances after
env.observation_layout             # named blocks: which channels/values mean what

reset(mask) samples parameters and states immediately. observe() renders as needed, so explicit render(mask) calls are optional. The training loop owns episode counters and chooses when to reset; the Gymnasium adapter below supplies same-step autoreset and time limits.

EnvConfig is the compile-time part (the JIT key); the runtime part is MultiEnvironment(..., parameters={"dynamics": {"dynamics": {"model": "x500", "mass": 2.1}}, "camera": {"fov": 90}}), a partial dict merged over the nominal parameters as the base of every drone (the specification's randomization ranges apply on top at reset). env.parameters returns drone zero's resolved nominal parameters, including its entry-specific overrides: "dynamics" is the l2f parameter set, where a "model" key loads an l2f registry entry (hyperdrone.dynamics.MODELS) before the rest is merged, and "camera" holds mount.pose (position and orientation, a (w, x, y, z) quaternion mapping the camera's FLU axes into the body frame) and the horizontal fov in degrees. env.instance_parameters() returns what each instance is using in the current episode (after randomization), and env.set_instance_parameters(overrides, mask) changes it until that instance's next reset (one dict for all selected instances or one per instance); only the drone itself varies per instance (dynamics.dynamics, dynamics.imu, camera).

Several drones fly in one environment through the drone table, MultiEnvironment(..., drones=[{"asset": x500, "parameters": {...}}, {"asset": crazyflie, "parameters": {"dynamics": {"dynamics": {"model": "crazyflie"}}}}, ...]): every instance samples one entry uniformly at each reset, independently of the scene its environment shows, so every scene meets every drone. An entry pairs the body/prop_* GLB rendered into the instance's cameras with a partial parameter dict merged over the shared constructor base (the per-drone keys dynamics.dynamics, dynamics.imu, camera); the same GLB may back several dynamics entries and distinct GLBs load once. EnvConfig(max_drones=...) is the table capacity (compile-time, default 1), env.drones returns the resolved native table, and the "drone" value in env.instance_parameters() indexes it. env.self_visible and env.max_drones report the compiled world's capabilities, including custom spec_header worlds. drone_asset=X is the one-entry table. The rigged assemblies of hyperdrone.examples.data.RIGGED_MODELS (arpl, crazyflie, crazyflie_brushless, soft, x500) pair with the l2f registry entries of the same names; each prop spins with the dynamics rotor at its position whatever rotor order the entry uses.

All environment semantics — reset, reward, termination, scene scheduling, observation composition — live on the C++ side (rl_tools::rl::environments::hyperdrone::MultiEnvironment<World>); the binding marshals tensors and nothing else, and a seeded rollout is pinned bit-exact against the C++ verbs by a golden test. The scenes argument is a directory of .glb scenes (sorted corpus) or a list of references — .glb paths, conta:HASH strings, or conta store manifest entries {"description": ..., "hash": ...} — used as the corpus in list order and partitioned across environments; conta references (scenes, drone_asset=, gate_asset=) resolve through the conta package into the shared content-addressed cache, so a config pins the exact corpus on any machine. Configuration follows the C++ extension ladder: preset= names the platform ("crazyflie", "x500_fpv" — SELF_VISIBLE, pass a body/prop_* GLB via drone_asset=; hyperdrone.env.Rotorcraft(assembly, rotor_positions=..., rotor_torque_directions=...) is the same rig as a render.Rig for manual composition: every prop_* part spins about its hub, and given the dynamics rotor layout (sim.model_parameters()) coordinate i is the phase of rotor i spinning against its reaction torque, so sim.state["rpm"] integrates straight into the coordinates; without rotors the props follow part order and the Quad-X convention, front-right CCW — and "x500_fpv_imu", the same platform stepping at IMU rate), task= names the wrapper ("target_frame", "moving_gate", "visual_inertial_localization"), n_agents= enables multi-agent, and EnvConfig(spec_header=...) pins an arbitrary C++ specification (a header defining hyperdrone_env_user::WORLD, hashed into the JIT key). A spec header can go beyond constants to a full user-authored task wrapper — verb overloads registering and moving entities, compiled reward, termination: hyperdrone/examples/Orbiter.ipynb walks through one, and its contract is pinned by tests/env/user_task_header.h. The environment's raytracing backend follows HYPERDRONE_RENDER_BACKEND like the renderer.

Manual composition of Sim and Renderer (no MDP — rendering research, data generation) remains first-class: see python -m hyperdrone.examples.drone_flythrough for the wiring. FreeSpaceSampler lives in hyperdrone.render — it rides the renderer's collision probes and is deterministic given a seed.

Visual-inertial localization

task="visual_inertial_localization" on preset="x500_fpv_imu" is the localization benchmark: fixed-length synchronized episodes stepping at IMU rate (env.dt), a camera frame every env.frame_stride steps, and the IMU sample of each step from env.observe_imu() (named by env.observation_layout_imu: accelerometer, gyroscope, frame age, new-frame flag). The task is autonomous: the RAPTOR autopilot built into it flies a waypoint route (its checkpoint is pinned by content hash and fetched through conta on first use), so action_dim is 0 and step() takes no actions. The privileged observation carries the current waypoint next to the dynamics state; the drone's own body is not rendered, matching the C++ benchmark harness. Resets must be all-or-none per environment.

config = EnvConfig(instances=2, cam_width=160, cam_height=120, preset="x500_fpv_imu", task="visual_inertial_localization")
env = MultiEnvironment(scenes, config=config, seed=0)
env.reset(); env.render(np.ones(env.total_instances, dtype=bool))
for step in range(env.episode_step_limit):
    if step > 0: env.render()
    if step % env.frame_stride == 0:
        frames = env.frames()          # the estimator's camera frame
    env.step()
    imu = env.observe_imu()            # the estimator's IMU sample for this step
    truth = env.observe_privileged()   # position / rotation matrix / velocities / waypoint

python -m hyperdrone.examples.visual_inertial_localization runs an episode and scores IMU dead reckoning against the ground truth, the same protocol as the C++ demo.

Gymnasium

from hyperdrone.gym import VectorEnv   # pip install "hyperdrone[gym]"
env = VectorEnv("scenes/", config=EnvConfig(instances=1024), seed=0)
observations, infos = env.reset()
observations, rewards, terminations, truncations, infos = env.step(actions)

The adapter resets instances that terminate or reach episode_step_limit within the same step. Returned observations start the new episode for those instances; no final observation is exposed. truncations flags time limits that are not terminations. The core packages never import gymnasium.

End-to-end example: python -m hyperdrone.examples.drone_flythrough; renderer benchmark: python -m hyperdrone.examples.benchmark (flag-compatible with the C++ benchmark counterpart).

Publishing

From the repository root, build and upload a release with:

interface/python/hyperdrone/scripts/publish.sh

Set the release version in pyproject.toml and hyperdrone/__init__.py first. The script uses the repository's .venv (creating it if needed), installs the build and upload tools, builds a bundled source archive and a py3-none-any wheel from that archive, checks both with Twine, and uploads both to PyPI. Authenticate through your existing Twine configuration (~/.pypirc, keyring, or TWINE_PASSWORD), or enter a PyPI API token when prompted.

Use interface/python/hyperdrone/scripts/publish.sh --build-only to build and check without uploading. Each run keeps its two artifacts in a fresh interface/python/hyperdrone/dist/release.* directory, so older builds are never included in an upload. Native components remain JIT-compiled on first use.

Tests

HYPERDRONE_RENDER_BACKEND=GENERIC .venv/bin/python -m pytest interface/python/hyperdrone/tests -v

tests/jit exercises the build infrastructure against a toy component in seconds; tests/test_architecture.py enforces the import DAG (render and dynamics never see each other); scripts/test_sdist.sh is the self-containment rot guard (build sdist → clean venv → vendored-tree render).

For labelled ProcTHOR rendering, preserve GLB root instances and select the output meaning:

scene = render.load_scene(procthor_scene_path(), rgb=False, preserve_instances=True)
renderer = render.Renderer(width=512, height=512, output="segmentation",
                           semantic_segmentation=True)  # False selects instance IDs
renderer.init(scene)
# Set cameras and render as usual.
labels = renderer.segmentation()  # uint32; renderer.segmentation_mode identifies the mode
classes = scene.segmentation_classes  # {class_id: name}, taxonomy: scene.segmentation_taxonomy

Semantic IDs come from AI2THOR-Hab's taxonomy: unknown is 0 (including Ceiling, which has no dataset category); background is 0xFFFFFFFF in both modes. Instance IDs identify scene instances: use scene.instance_object(id) then scene.object_name(object_index) for static hits. Overlay instance IDs use the renderer's global slot layout. Semantic IDs must never be used as object indices. Loading remains welded by default for RGB; semantic rendering rejects a welded object containing multiple classes. Assemblies and asset-pool copies retain labels and the taxonomy. Composing conflicting taxonomies requires an explicit remapping before semantic rendering.

Assign labels before Renderer.init(): object.segmentation_class = id, scene.set_object_segmentation_class(index, id), assembly.set_segmentation_class(index, id), or pool.set_object_segmentation_class(asset, object_index, id) retains the current taxonomy and explicitly makes that object's class valid, including welded objects. Assignment rejects IDs absent from that taxonomy. object.assign_segmentation_class(id, taxonomy="custom", classes={0: "Unknown", 7: "Furniture"}) replaces the taxonomy; omitting taxonomy/classes selects unnamed numeric IDs.

Objects, scenes, assemblies and pools expose remap_segmentation_classes(mapping, taxonomy="custom", classes={...}); pools also expose remap_asset_segmentation_classes(asset, mapping, ...). Omitting taxonomy/classes selects unnamed numeric IDs. Mappings must cover every used class, and validation failures leave labels unchanged. Remapping preserves valid class boundaries and rejects mixed-class welded objects; explicitly assign one class first if collapsing their labels is intentional. Remap conflicting sources separately before composition. The C++ equivalents are device-first assign_segmentation_class and remap_segmentation_classes in rendering/segmentation_cpu.h, with a shared immutable target taxonomy (or nullptr for unnamed IDs).

Articulated assets

render.Rig is a rigid-link tree: world_link = world_parent @ origin @ joint(q) and world_geometry = world_link @ offset, with links declared parent before child (parent -1 is the base pose) and joints fixed, revolute (radians about axis), prismatic (metres along axis) or spherical (wxyz quaternion). rig.coordinate_slices locates each link's coordinates, rig.default_coordinates is the neutral state. Rig.from_assembly(assembly, joints={"slider": ("prismatic", (0, 1, 0))}) builds one link per part at its rest frame and articulates parts by name; load_assembly(path, part_nodes=[...]) keeps nested GLB nodes as independent parts (an enclosing node first) and records their parent for that. Procedural assets grow with ObjectAssembly.add_part(object, transform). The drone rig the environment renders is hyperdrone.env.Rotorcraft, a Rig built from the body/prop_* convention (see the RL environment section).

renderer.set_transform(overlay, placement, body, rig=rig, coordinates=phases)
renderer.set_transform_pair(overlay, placement, body_open, body_close, rig=rig,
                            coordinates_open=phases_open, coordinates_close=phases_close)
renderer.update()
renderer.render()

With a rig, the rigid verbs take the rig's base pose and joint coordinates (default: the neutral state) and place the parts at its geometry-to-world poses; the pair form samples the shutter (the base follows the constant-twist screw, the coordinates interpolate before the kinematics, so a revolute joint keeps every revolution). The same poses are available as arrays: rig.evaluate(base, q) is (num_parts, 3, 4) for set_transforms, rig.sample(base_open, base_close, q_open, q_close, renderer.motion_blur_samples) the (samples, parts, 3, 4) motion plus the shutter-close poses for set_motion_transforms. Batches of identical assets, one per overlay, submit the whole tensor at once with set_all_transforms(poses) of shape (num_overlays, max_overlay_instances, 3, 4) (and set_all_motion_transforms); rig.evaluate accepts batched (B, 3, 4) bases for that. On OptiX every transform verb also takes CUDA tensors (stream= orders the copy after the producer), and renderer.transforms_dlpack() / transforms_motion_dlpack() expose the live input tensors for in-place producers ordered against renderer.render_stream; publish pending host writes with update() before writing them. python -m hyperdrone.examples.articulation renders a spinning hinge and a sliding block with motion blur; hyperdrone/examples/Custom.ipynb flies the x500 with spinning props.