Skip to content

Indoor robots (IMU + wheel odometry, no GPS)

For any ground robot running indoors or in GPS-denied environments with wheel encoders as the primary motion source.

Quick start

ros2 launch fusioncore_ros fusioncore.launch.py \
  fusioncore_config:=$(ros2 pkg prefix fusioncore_ros)/share/fusioncore_ros/config/wheels_indoor.yaml \
  --ros-args \
  -r /odom/wheels:=/your/wheel/odom/topic

Common wheel odom topic remaps:

Robot / controller Topic Remap
ROS 2 Control diff_drive /diff_controller/odom -r /odom/wheels:=/diff_controller/odom
Turtlebot3 /odom -r /odom/wheels:=/odom
Nav2 default /odom -r /odom/wheels:=/odom
Already /odom/wheels : no remap needed

What FusionCore gives you indoors

Without GPS, FusionCore gives you clean local odometry: not global position. It starts at the origin and builds from there. What it adds over raw wheel odom:

  • IMU gyro + accel bias estimated continuously as filter states → less heading drift
  • Chi-squared outlier rejection on wheel odom → encoder spikes don't corrupt the state
  • ZUPT (zero-velocity updates) → drift suppressed while stationary
  • Adaptive noise → covariance tracks actual sensor behavior, not guessed config values

For global positioning and loop closure, pair with a SLAM system.

SLAM integration

FusionCore and SLAM divide the TF tree cleanly:

map → odom        ← slam_toolbox or RTABMAP
odom → base_link  ← FusionCore

# FusionCore (odometry)
ros2 launch fusioncore_ros fusioncore.launch.py \
  fusioncore_config:=wheels_indoor.yaml \
  --ros-args -r /odom/wheels:=/diff_controller/odom

# SLAM
ros2 launch slam_toolbox online_async_launch.py

Adding LiDAR ICP as a second velocity source

If your wheel odom is unreliable (slippery floors, wheel slip) or you want cross-validation, add LiDAR ICP alongside wheel odom. Uncomment in wheels_indoor.yaml:

encoder2.topic: "/kiss/odometry"    # KISS-ICP
# encoder2.topic: "/icp_odom"       # rtabmap icp_odometry

FusionCore fuses both independently. If wheel odom has a spike, the chi-squared gate rejects it: ICP keeps the estimate grounded.

If you have no wheel odom at all and only LiDAR ICP, use icp_indoor.yaml instead.

One thing to get right: startup bias init

Indoors, without GPS corrections, IMU bias matters. wheels_indoor.yaml has this enabled by default:

init.stationary_window: 2.0

Keeps the robot still for 2 seconds at launch, estimates accelerometer bias from the mean readings, then starts. Reduces startup drift from ~10 cm to under 1 cm. Falls back automatically if the robot moves during the window.