Signal types¶
from nexus_uav import SimTime, PoseTwist, PositionGoal, Waypoints, ReferenceTrajectory
from nexus_uav import ImuSample, MagSample, BaroSample, GpsSample, Image, PointCloud
A signal carries one of these types between components, and a
component's stage declares its signals with them: Signal("estimate", PoseTwist, shape=(1,)). A
Warp struct lives on the device, and a host stage reads it as an instance of its type,
read().accel. The controls a controller writes are a plain device row of wp.float32, one per
actuator, so they have no type here.
dataclass
¶
SimTime(time_ns: int = 0, step_index: int = 0)
Sim-time and step index for the current tick.
The single authoritative time source for HIL_* MAVLink timestamps and the sensors' due times,
so every component shares one consistent clock derived from the simulation. The time is a whole
number of nanoseconds, as gz-sim keeps it, so a time that falls on a tick compares exactly; a
stage reads it in seconds.
class-attribute
instance-attribute
¶
time_ns: int = 0
Elapsed simulation time since reset, in whole nanoseconds.
class-attribute
instance-attribute
¶
step_index: int = 0
Zero-based index of the current simulation step (integer tick counter).
property
¶
time_usec: int
Simulation time in integer microseconds, the HIL_* timestamp unit.
The pose and twist of one body in world axes: what the plant writes per body, as the signal
body/<label> of shape (1,), and the estimate an estimator writes, as the signal estimate,
which the guidance and the controllers read.
dataclass
¶
PositionGoal(
pos: tuple[float, float, float],
yaw: float | None = None,
)
A single move-to / hold goal in the world frame, Newton FLU / Z-up. Consumed by the
state-feedback controllers, policy and pid: the guidance feeds one PositionGoal at a time and
sequences a mission by advancing it on arrival; the controller is goal-relative, so each is a
fresh single-goal problem. yaw is the optional heading [rad]; None = don't command yaw.
The signal setpoint carries it on the device as one wp.vec3 of the position, since no
controller reads yaw.
dataclass
¶
Waypoints(
points: list[tuple[float, float, float]] = list(),
)
An ordered list of world-frame positions the controller holds and advances internally.
Distinct from a mission the guidance sequences with PositionGoal: here the whole path is the
setpoint.
class-attribute
instance-attribute
¶
points: list[tuple[float, float, float]] = field(
default_factory=list
)
dataclass
¶
ReferenceTrajectory(reference: Any = None)
A full flat-state reference over time, for the acados NMPC: the guidance plans it, ruckig →
differential-flatness, from waypoints and feeds it to the controller, which tracks it. The
payload is a queryable reference object, for example the guidance's FlatnessReference, the
controller samples per horizon node, kept opaque here so core imports no controller backend.
The sample of an Inertial Measurement Unit (IMU), in the axes of its mount: body
Forward Right Down (FRD) on an unturned mount. The signal imu carries it, of shape (1,).
A barometer's sample. The signal baro carries it, of shape (1,).
A Global Positioning System (GPS) receiver's sample, on WGS84. The signal gps carries it, of shape
(1,). Latitude, longitude and altitude are 64-bit: a 32-bit latitude loses about a metre.
instance-attribute
¶
fix_type: int32
The fix, as MAVLink's GPS_FIX_TYPE numbers it: 3 is a 3D fix.
dataclass
¶
Image(time: float, pixels: ndarray)
dataclass
¶
PointCloud(time: float, points: ndarray)