A Reactive View Planner For Mobile Robots

This codebase contains the core implementation of the First Look Inspection Planner developed as part of the research pursued in my doctoral thesis On Adaptive and Scene-Aware Inspection Autonomy.

The codebase has been tested on Quadrupedal robots (SPOT), Ground robots (Husky) and Aerial robots (Shafter). The method has been extensively validated across static and dynamic environment condition as well as in a priori known and unknown environments. The work has been validated using both 3D LiDAR and RGBD point-clouds as inputs. For better performance, use of LiDAR points is advised in combination with the voxel grid filter. The codebase was mostly evaluated in ROS 1 and has been ported to ROS 2.

The method plans over instantaneous 3D point-cloud and odometry information with configurable orientation of the onboard sensor suite.

Kindly cite the following paper if you find the codebase useful for your work.

@phdthesis{KottayamViswanathan2026,
  author       = {Kottayam Viswanathan, Vignesh},
  title        = {On Adaptive and Scene-Aware Inspection Autonomy in Challenging Environments},
  school       = {Luleå University of Technology},
  year         = {2026},
  isbn         = {978-91-8048-993-5},
  url          = {https://urn.kb.se/resolve?urn=urn:nbn:se:ltu:diva-116469}
}

Ackowledgements

This work was made possible in collaboration with the amazing team at Robotics and Artificial subject at Lulea University of Technology. Visit www.fieldrobotics.eu for more information.

License

MIT- modify it and adapt for your use-case.

Installation Instructions (Ubuntu 22.04 and ROS2 Humble)

cd colcon_workspaces
mkdir -p inspection_ws/src
cd inspection_ws/src
git clone git@github.com:Astrabrat/FLIP-First-Look-Inspection-Planner.git

pip install ttictoc loguru

Build ROS2 package

colcon build --symlink-install --cmake-args -DCMAKE_BUILD_TYPE=Release

Deployment Instructions

Run ros2 launch inspection_planner planner.launch.py team:=kReal robot:=husky to start the inspection planner

Open another terminal, source the inspection_ws and enter the following

Run ros2 launch inspection_planner voxel_grid_simple.launch.py team:=kReal robot:=husky to launch the voxel grid filter

The input params robot and team define the robot namespace and type of deployements, e.g. simulation or real robots. This launches the appropriate config files. Change it to suit your needs.

Check ODOM and PCL topics. Once ready to start planner, send the following service,

ros2 service call /<robot_namespace>/initialize_inspection std_srvs/srv/Trigger {}
# Example
ros2 service call /husky/initialize_inspection std_srvs/srv/Trigger {}

Else, run pip install tmuxinator and launch the stack with,

tmux start-server \; source-file src/FLIP-First-Look-Inspection-Planner/utilities/config/inspection.tmux.conf \; attach-session -t inspection

Performance Monitoring

# InspectionPerformance.msg

# Standard ROS header
std_msgs/Header header

std_msgs/Float64 view_planning_time

std_msgs/String current_mission_status

# standard performance metrics
std_msgs/Float64 maintained_distance
std_msgs/Float64 desired_distance
std_msgs/Float64 view_quality

Configuration

To configure topics and mission parameters, modify /inspection_planner/config/rosparams_exp.yaml file. A brief overview of the parameters are provided below

Parameters

General

Name	Type	Default	Description
name	string	"husky"	Robot identifier.
world_frame	string	"world"	Global reference frame.
base_link_frame	string	"husky/base_link"	Robot base frame.
sensor_frame	string	"husky/base_link"	Sensor reference frame.

Platform & Modes

Name	Type	Default	Description
platform_modality	int	0	Platform type (0: ground, 1: aerial).
run_mode	int	0	Mode (0: onboard, 1: debug).
sensor_modality	int	0	Sensor type (0: camera).

Sensor Configuration

Name	Type	Default	Description
fov	float[2]	[69.4, 45.0]	Horizontal/vertical field of view (deg).
aspect_ratio	float	1.33	Camera aspect ratio (width/height).
sensor_rotation	float[3]	[0.0, 0.0, -1.57]	Rotation (roll, pitch, yaw in rad).
sensor_translation	float[3]	[0.0, 0.0, 0.0]	Translation offset (m).

Control & Planning

Name	Type	Default	Description
rate_controller	float	1.0	Control loop frequency (Hz).
cmd_pos_upd	float	25.0	Position command update threshold (m).
cmd_yaw_upd	float	3.14	Yaw update threshold (rad).
prediction_horizon	int	10	Prediction horizon (steps).
confidence_horizon	int	5	Confidence horizon (steps).

Inspection Settings

Name	Type	Default	Description
inspection_distance	float	1.5	Target distance to object (m).
inspection_height	float	1.0	Target inspection height (m).
photogrammetric_params	float[2]	[0.7, 0.01]	Photogrammetry tuning parameters.

ROS Topics

Name	Type	Default	Description
odom_topic	string	"odometry/imu"	Odometry input topic.
pcl_topic	string	"filtered_pointcloud"	Point cloud input topic.
sbl_topic	string	"sbl/data"	Altitude input topic (for aerial robots).
reference_pose	string	"command/pose"	Reference pose topic.
inspection_performance	string	"inspection_planner/results/inspection_performance"	Performance metrics output.
tracked_path	string	"inspection_planner/results/tracked_path"	Executed trajectory.
predicted_path	string	"inspection_planner/results/predicted_path"	Predicted trajectory.
maintained_distance	string	"inspection_planner/results/maintained_distance"	Distance tracking output.
cropped_points	string	"inspection_planner/results/cropped_points"	Filtered point cloud output.
cam_frustum	string	"inspection_planner/general/camera_frustum"	Camera frustum visualization.

Note

• Adjust the sensor_rotation parameters as needed (currently supported).
• sensor_translation parameters are defined but not yet supported.
• Configure the voxel grid filter by setting the subsampling resolutions for the X, Y, and Z axes.
• There are some open parameters (defined but not used) for future proper improvements.

TODO

1. Remove wildcard imports for control and debugging
2. Integrate sensor_translation effect properly into view-planning
3. Address open parameters