-
Notifications
You must be signed in to change notification settings - Fork 132
Open
Description
I'm using the rolling branch of 'fuse' to fuse the output of lidar scan matching (absolute x, y and yaw) and wheel odometry (linear x and angular yaw velocity). I observe the estimated output lagging followed by overshoot and eventual correction. This causes the stationary objects (walls, parked cars etc.) to appear moving around the robot in the fixed frame (map). I've tried to capture this in the screen recording linked below.
I also recorded a bagfile while recording the video.
My configuration is as follows:
fuse_map:
ros__parameters:
# Fixed-lag smoother configuration
optimization_frequency: 20.0
transaction_timeout: 0.1
lag_duration: 1.0
motion_models:
unicycle_motion_model:
type: fuse_models::Unicycle2D
unicycle_motion_model:
# x y yaw vx vy vyaw ax ay
process_noise_diagonal: [0.100, 0.100, 0.100, 0.100, 0.100, 0.100, 0.1, 0.1]
sensor_models:
initial_localization_sensor:
type: fuse_models::Unicycle2DIgnition
motion_models: [unicycle_motion_model]
ignition: true
wheel_odometry:
type: fuse_models::Odometry2D
motion_models: [unicycle_motion_model]
ndt_pose:
type: fuse_models::Pose2D
motion_models: [unicycle_motion_model]
initial_localization_sensor:
publish_on_startup: true
# x y yaw vx vy vyaw ax ay
initial_state: [0.000, 0.000, 0.000, 0.000, 0.000, 0.000, 0.000, 0.000]
initial_sigma: [0.100, 0.100, 0.100, 0.100, 0.100, 0.100, 0.100, 0.100]
wheel_odometry:
topic: 'wheel_odom'
twist_target_frame: 'base_link'
linear_velocity_dimensions: ['x', 'y']
angular_velocity_dimensions: ['yaw']
ndt_pose:
topic: 'pcl_pose'
position_dimensions: ['x', 'y']
orientation_dimensions: ['yaw']
publishers:
filtered_publisher:
type: fuse_models::Odometry2DPublisher
filtered_publisher:
topic: 'odometry/global'
base_link_frame_id: 'base_link'
odom_frame_id: 'odom'
map_frame_id: 'map'
world_frame_id: 'map'
publish_tf: true
publish_frequency: 20.0
I have also tried setting optimization_frequency and publish_frequency to 10, and lag_duration to 0.5 but it did not help.
Metadata
Metadata
Assignees
Labels
No labels
