# Context & Hardware Setup

**URL:** <https://discourse.openrobotics.org/t/context-hardware-setup/55356>\
**Category:** ROS General\
**Tags:** ros2, ros, humble, jazzy\
**Created:** [June 8, 2026, 4:26pm UTC](https://discourse.openrobotics.org/t/context-hardware-setup/55356 "2026-06-08T16:26:45Z")\
**Posts on this page:** 1\
**Page:** 1

<div class="post-metadata">

**Author:** ![Ali\_Altin](https://sea2.discourse-cdn.com/flex022/user_avatar/discourse.openrobotics.org/ali_altin/32/24472_2.png) [@Ali\_Altin](https://discourse.openrobotics.org/u/Ali_Altin)\
**Post date:** [June 8, 2026, 4:26pm UTC](https://discourse.openrobotics.org/t/context-hardware-setup/55356/1 "2026-06-08T16:26:46Z")

</div>

I am developing an Autonomous Mobile Robot (AMR) designed to operate inside a dairy cattle barn. The locomotion system suffers from high wheel slippage due to the environment’s muddy, uneven, and unpaved terrain.

My current sensor suite consists of:

- **Wheel Encoders** (Differential drive odometry)

- **IMU** (Providing angular velocity and absolute orientation)

- **Ultra-Wideband (UWB)** Positioning System (1 Tag on the robot, multiple Anchors around the barn)

- _(Upcoming)_ Range sensors for obstacle avoidance (cattle detection).

For state estimation, I am using the `robot_localization` package running a **Dual EKF** setup to broadcast the `map -> odom -> base_footprint` transform tree.

## The Problem

I am experiencing significant localization inaccuracies, specifically with the robot’s heading (yaw) and absolute positioning.

1. **Heading / Yaw Errors:** The estimated orientation is frequently wrong or drifts rapidly. The barn structure contains dense steel pillars (headlocks and support frames) which I suspect are causing massive magnetic disturbances to the IMU’s magnetometer.

2. **UWB Inaccuracy & Jumps:** Initially, the UWB update rate was 1 Hz, which caused huge position jumps. I managed to optimize the anchor-tag communication to achieve around **10 Hz**. While the frequency issue is resolved, I still observe positional errors between **10 cm to 40 cm** , which drastically degrade especially when the robot approaches the steel pillars or when cattle walk between the tag and the anchors.

## Current Configuration & Attempts

- The first EKF instance fuses wheel odometry and IMU (angular velocity) to output continuous `odom -> base_footprint`.

- The second EKF instance fuses the `odom` data and absolute UWB positions to output `map -> odom`.

- I have noticed that high multipath reflections and Non-Line-of-Sight (NLOS) conditions caused by steel structures and livestock are heavily corrupting the UWB ranges.

## Questions for the Community

Given the harsh environment (muddy terrain, heavy metal interference, dynamic obstacles like cows), I would highly appreciate any advice or architectural suggestions:

1. **How should I handle the IMU in a high-metal environment?** Should I completely disable the magnetometer fusion in `robot_localization` and rely solely on `differential_drive` odometry + IMU angular velocity (z-axis gyro) for heading, even if it drifts over time?

2. **How can I mitigate UWB multipath/NLOS errors in EKF?** Are there proven outlier rejection methods or thresholding techniques within `robot_localization` (like using the `covariance` matrix dynamically or `mahalanobis_distance` filtering) to ignore bad UWB fixes near the steel pillars?

3. **Alternative Fusion Strategies:** Has anyone successfully implemented robust UWB-based localization in similar agricultural setups? Would moving towards an Unscented Kalman Filter (UKF) or a factor-graph-based optimization approach (like Ceres or GTSAM) yield better results than standard Dual EKF?

Thank you in advance for your insights!

My ekf configuration :

```auto
odometry_ekf_filter_node:
    ros__parameters:
        sensor_timeout: 0.1
# The frequency, in Hz, at which the filter will output a position estimate. Note that the filter will not begin
# computation until it receives at least one message from one of the inputs. It will then run continuously at the
# frequency specified here, regardless of whether it receives more measurements. Defaults to 30 if unspecified.
        frequency: 30.0
        permit_corrected_publication: false
# ekf_localization_node and ukf_localization_node both use a 3D omnidirectional motion model. If this parameter is
# set to true, no 3D information will be used in your state estimate. Use this if you are operating in a planar
# environment and want to ignore the effect of small variations in the ground plane that might otherwise be detected
# by, for example, an IMU. Defaults to false if unspecified.
        two_d_mode: true

# Use this parameter to provide an offset to the transform generated by ekf_localization_node. This can be used for
# future dating the transform, which is required for interaction with some other packages. Defaults to 0.0 if
# unspecified.
        transform_time_offset: 0.0

# Use this parameter to provide specify how long the tf listener should wait for a transform to become available.
# Defaults to 0.0 if unspecified.
        transform_timeout: 0.05

# If you're having trouble, try setting this to true, and then echo the /diagnostics_agg topic to see if the node is
# unhappy with any settings or data.
        print_diagnostics: false

# Whether to publish the acceleration state. Defaults to false if unspecified.
        publish_acceleration: false

# Whether to broadcast the transformation over the /tf topic. Defaults to true if unspecified.
        publish_tf: true
        # initial_state: [2.0, -8.0, 0.0,
        # 0.0, 0.0, 0.0,
        # 0.0, 0.0, 0.0,
        # 0.0, 0.0, 0.0,
        # 0.0, 0.0, 0.0]
# 1. Set the map_frame, odom_frame, and base_link frames to the appropriate frame names for your system.
# 1a. If your system does not have a map_frame, just remove it, and make sure "world_frame" is set to the value of odom_frame.
# 2. If you are fusing continuous position data such as wheel encoder odometry, visual odometry, or IMU data, set "world_frame"
# to your odom_frame value. This is the default behavior for robot_localization's state estimation nodes.
# 3. If you are fusing global absolute position data that is subject to discrete jumps (e.g., GPS or position updates from landmark
# observations) then:
# 3a. Set your "world_frame" to your map_frame value
# 3b. MAKE SURE something else is generating the odom->base_link transform. Note that this can even be another state estimation node
# from robot_localization! However, that instance should *not* fuse the global data.
        #map_frame: map # Defaults to "map" if unspecified
        odom_frame: odom # Defaults to "odom" if unspecified
        base_link_frame: base_footprint # Defaults to "base_link" if unspecified
        world_frame: odom # Defaults to the value of odom_frame if unspecified

        odom0: /wheel_odom
        # x, y pozisyon differential modda fuse ediliyor.
        # Twist (vx) kapalı - twist verisine güvenilmiyor.
        # differential: true -> mutlak pose değil, adım adım delta alınıyor,
        # böylece birikimli odometri hatası direkt EKF'e yansımaz.
        odom0_config: [true, true, false,
                      false, false, false,
                      false, false, false,
                      false, false, false,
                      false, false, false]
        odom0_differential: true
        odom0_relative: false
        odom0_queue_size: 10
        odom0_pose_rejection_threshold: 5.0
        odom0_twist_rejection_threshold: 5.0

        imu0: /witmotion/imu
        imu0_config: [false, false, false,
                      false, false, false, # absolute yaw KAPALI — motor EMI manyetometreyi bozuyor, gradual drift yaratıyor
                      false, false, false,
                      false, false, true, # yaw_rate (jiroskop) — SLAM map→odom global yaw'ı düzeltiyor
                      false, false, false]
        imu0_differential: false
        imu0_relative: false
        imu0_queue_size: 10
        # EMI koruması: 10.0'dan 3.0'a düşürüldü.
        # Motorlardan kaynaklanan magnetik spike'lar bu eşiği aşarsa elenir.
        # Eğer yaw sürekli reddediliyorsa (diagnostics'te görünür) değeri
        # biraz artır (4.0-5.0) ya da imu0_config[5]'i false yap.
        imu0_pose_rejection_threshold: 3.0
        imu0_twist_rejection_threshold: 5.0
        imu0_linear_acceleration_rejection_threshold: 0.8
        imu0_remove_gravitational_acceleration: true

```

```auto
ekf_filter_node:
  ros__parameters:
    frequency: 10.0
    sensor_timeout: 0.5
    two_d_mode: true # Set false for 3D navigation (drones, etc.)
    transform_time_offset: 0.0
    transform_timeout: 0.0
    print_diagnostics: true
    debug: false
    publish_tf: false

    map_frame: map
    odom_frame: odom
    base_link_frame: base_footprint
    world_frame: map
    # [x, y, z, roll, pitch, yaw, vx, vy, vz, vroll, vpitch, vyaw, ax, ay, az]
    # initial_state: [2.55, -7.4, 0.0,
    # 0.0, 0.0, 0.0,
    # 0.0, 0.0, 0.0,
    # 0.0, 0.0, 0.0,
    # 0.0, 0.0, 0.0]
    # -----------------------------------------------------------------------
    # Sensor 0: Wheel Odometry
    # -----------------------------------------------------------------------
    # Fuse: vx, vy, vyaw (velocities from wheel encoders)
    # [x, y, z, roll, pitch, yaw, vx, vy, vz, vroll, vpitch, vyaw, ax, ay, az]
    odom0: /wheel_odom
    # x, y pozisyon differential modda fuse ediliyor.
    # Twist (vx) kapalı - twist verisine güvenilmiyor.
    # differential: true -> mutlak pose değil, adım adım delta alınıyor,
    # böylece birikimli odometri hatası direkt EKF'e yansımaz.
    odom0_config: [true, true, false,
                  false, false, false,
                  false, false, false,
                  false, false, false,
                  false, false, false]
    odom0_differential: true
    odom0_relative: false
    odom0_queue_size: 10
    odom0_pose_rejection_threshold: 5.0
    odom0_twist_rejection_threshold: 5.0

    imu0: /witmotion/imu
    imu0_config: [false, false, false,
                  false, false, false, # absolute yaw KAPALI — motor EMI manyetometreyi bozuyor, gradual drift yaratıyor
                  false, false, false,
                  false, false, true, # yaw_rate (jiroskop) — SLAM map→odom global yaw'ı düzeltiyor
                  false, false, false]
    imu0_differential: false
    imu0_relative: false
    imu0_queue_size: 10
    # EMI koruması: 10.0'dan 3.0'a düşürüldü.
    # Motorlardan kaynaklanan magnetik spike'lar bu eşiği aşarsa elenir.
    # Eğer yaw sürekli reddediliyorsa (diagnostics'te görünür) değeri
    # biraz artır (4.0-5.0) ya da imu0_config[5]'i false yap.
    imu0_pose_rejection_threshold: 3.0
    imu0_twist_rejection_threshold: 0.8
    imu0_linear_acceleration_rejection_threshold: 0.8
    imu0_remove_gravitational_acceleration: true

    # -----------------------------------------------------------------------
    # Sensor 1: UWB Trilateration Position
    # -----------------------------------------------------------------------
    # Fuse: x, y, z (absolute position from trilateration)
    # [x, y, z, roll, pitch, yaw, vx, vy, vz, vroll, vpitch, vyaw, ax, ay, az]
    pose0: /uwb/pose
    pose0_config: [true, true, false,
                   false, false, false,
                   false, false, false,
                   false, false, false,
                   false, false, false]
    pose0_differential: false
    pose0_queue_size: 10
    pose0_rejection_threshold: 1.0 # Reject UWB jumps > 2m (NLOS protection)

```