IMU Sensor#

The IMUSensor (Inertial Measurement Unit) provides accelerometer, gyroscope, and magnetometer readings for robot state estimation.

Overview#

An IMU sensor measures:

Linear acceleration: 3D acceleration including gravity
Angular velocity: 3D rotational velocity
Magnetic field: 3D magnetic field vector

Usage#

import genesis as gs

gs.init()
scene = gs.Scene()
robot = scene.add_entity(gs.morphs.URDF(file="quadruped.urdf"))
base = robot.get_link("base_link")

# Add IMU to robot base
imu = scene.add_sensor(
    gs.sensors.IMU(
        entity_idx=robot.idx,
        link_idx_local=base.idx_local,
        pos_offset=(0.0, 0.0, 0.15),
    )
)

scene.build()

# Simulation loop
for i in range(1000):
    scene.step()

    # Get IMU readings (IMUReturnType NamedTuple)
    data = imu.read()
    accel = data.lin_acc  # ([n_envs,] 3) acceleration in m/s^2
    gyro = data.ang_vel   # ([n_envs,] 3) angular velocity in rad/s
    mag = data.mag        # ([n_envs,] 3) magnetic field in Tesla

Configuration#

gs.sensors.IMU(
    # Attachment (inherited from RigidSensorOptionsMixin)
    entity_idx=robot.idx,             # Global entity index
    link_idx_local=base.idx_local,    # Local link index
    pos_offset=(0.0, 0.0, 0.0),      # Position offset from link frame
    euler_offset=(0.0, 0.0, 0.0),    # Rotation offset from link frame (degrees)

    # Accelerometer parameters
    acc_noise=(0.01, 0.01, 0.01),              # White noise std per axis (m/s^2)
    acc_random_walk=(0.001, 0.001, 0.001),     # Bias drift std per axis (m/s^3)
    acc_bias=(0.0, 0.0, 0.0),                  # Constant additive bias per axis
    acc_cross_axis_coupling=0.0,               # Cross-axis misalignment
    acc_resolution=0.0,                        # Measurement resolution (0 = no quantization)

    # Gyroscope parameters
    gyro_noise=(0.01, 0.01, 0.01),             # White noise std per axis (rad/s)
    gyro_random_walk=(0.001, 0.001, 0.001),    # Bias drift std per axis (rad/s^2)
    gyro_bias=(0.0, 0.0, 0.0),                # Constant additive bias per axis
    gyro_cross_axis_coupling=0.0,              # Cross-axis misalignment
    gyro_resolution=0.0,                       # Measurement resolution (0 = no quantization)

    # Magnetometer parameters
    mag_noise=(0.0, 0.0, 0.0),                # White noise std per axis
    mag_random_walk=(0.0, 0.0, 0.0),          # Bias drift std per axis
    mag_bias=(0.0, 0.0, 0.0),                 # Constant additive bias per axis
    mag_cross_axis_coupling=0.0,               # Cross-axis misalignment
    mag_resolution=0.0,                        # Measurement resolution (0 = no quantization)

    # Timing
    delay=0.0,                                 # Read delay in seconds
    jitter=0.0,                                # Random delay jitter in seconds

    draw_debug=True,
)

Output Format#

read() and read_ground_truth() both return an IMUReturnType NamedTuple:

Field	Type	Shape	Description
`lin_acc`	`torch.Tensor` (float32)	`([n_envs,] 3)`	Linear acceleration in local sensor frame (m/s^2)
`ang_vel`	`torch.Tensor` (float32)	`([n_envs,] 3)`	Angular velocity in local sensor frame (rad/s)
`mag`	`torch.Tensor` (float32)	`([n_envs,] 3)`	Magnetic field vector in local sensor frame (Tesla)

read() applies noise, bias, random walk, and cross-axis coupling if configured. read_ground_truth() returns noiseless values.

Noise Modeling#

The IMU supports realistic noise modeling based on Allan variance parameters:

Accelerometer Noise#

Parameter	Description	Typical Value
`acc_noise`	White noise std	0.001-0.01 m/s^2
`acc_random_walk`	Bias drift std	0.0001-0.001 m/s^3

Gyroscope Noise#

Parameter	Description	Typical Value
`gyro_noise`	White noise std	0.0001-0.001 rad/s
`gyro_random_walk`	Bias drift std	0.00001-0.0001 rad/s^2

Example: Quadruped State Estimation#

import genesis as gs
import numpy as np

gs.init()
scene = gs.Scene()
quadruped = scene.add_entity(gs.morphs.URDF(file="go2.urdf"))
base = quadruped.get_link("base")

# Add IMU with realistic noise
imu = scene.add_sensor(
    gs.sensors.IMU(
        entity_idx=quadruped.idx,
        link_idx_local=base.idx_local,
    )
)

scene.build()

# State estimation loop
import torch
velocity_estimate = torch.zeros(3, device=gs.device)
dt = scene.dt

for i in range(1000):
    scene.step()

    data = imu.read()

    # Simple integration (real systems use Kalman filtering)
    velocity_estimate += data.lin_acc * dt

API Reference#

See Also#

Sensors - Sensor overview
Contact Sensors - Contact force sensing