Skip to content

README.md

Latest commit

 

History

History
121 lines (83 loc) · 7.2 KB

README.md

File metadata and controls

121 lines (83 loc) · 7.2 KB
Raw

baseline_mission

Boustrophedon (lawnmower) coverage baseline for the VISTA UUV mission.

This is the non-adaptive comparison baseline for the NBV mission (demo_behaviors). It runs an exhaustive lawnmower survey over a rectangular ROI: a single standalone action client computes the coverage path in-file and dispatches the waypoints one at a time as PoseToPose goals to the shared Dubins action server. No behavior tree, no sampling/scoring servers.

It reuses the same simulator, Dubins action server, and sensor model as the NBV mission, so the two are directly comparable.

Quick Start

Build

cd ~/projects/rosmosis_ws
colcon build --symlink-install --packages-select baseline_mission
source install/setup.bash

(With --symlink-install, later edits to the node, launch file, and configs are picked up without rebuilding; a rebuild is only needed when adding files or changing setup.py/package.xml.)

Launch

ros2 launch baseline_mission baseline_mission_launch.py environment:=env_50x50_centroidBox

This starts:

  • vista_sim (simulator, Dubins action server, sensor model, RViz, static TFs)
  • boustrophedon_run (the lawnmower action client)
  • rqt_console (filterable log viewer) and rqt_graph (node/topic topology) — only when debug_gui:=true
  • ros2 bag record (only when record:=true; see Recording a run)

The client self-gates: it waits for the ned<-map TF (to resolve its survey depth from clearance) and for the pose_to_pose action server, so no launch-side delay is needed.

How it works

boustrophedon_run.py is a plain rclpy action client (not a BT node):

  1. Resolve depth from the TF — looks up ned<-map, reads seafloor_depth = t.z, and sets survey_depth = seafloor_depth - clearance so the vehicle flies a fixed clearance above the seabed.
  2. Build the lawnmower — rows run along north, lanes step along east; alternating direction (boustrophedon). Each lane emits only its two endpoints; the Dubins server plans the straight run between them and the U-turn to the next lane.
  3. Dispatch sequentially — send one PoseToPose goal, block until the result, send the next. A near-miss at a lane end (server returns "Missed") is treated as success, so an overshoot never aborts the survey.

Geometry

The forward-looking sonar footprint sets the lane layout. With the sensor at clearance z above the floor:

swath_cross  = 2 * z * tan(HFOV / 2)       # cross-track footprint width
lane_spacing = swath_cross / 2             # 50% side overlap
row_extension = z * (tan(mount + VFOV/2) - tan(mount - VFOV/2))   # along-track "running start"

The mount angle does not affect the swath/lane spacing (the FOV does); it only sets the along-track running start by which each row is extended past the ROI edges, so the swath fully sweeps the boundaries before the vehicle turns.

Frame convention

Poses are in NED (matches the Dubins action server): position.x = north, position.y = east, position.z = depth (down +); yaw = 0 heads north, yaw = pi heads south. The vehicle spawns at Eta(-5, 0, 0, yaw=0) (shared with the NBV mission), just south of the ROI on the first lane.

Parameters

Launch arguments (CLI-overridable, param:=value)

Parameter Default Description
environment environment_basic Environment yaml name (no extension) from sensor_model/config/
start_rviz true Start RViz with the simulation
drift_velocity 0.25 Idle drift velocity when not navigating (m/s)
constant_velocity 0.5 Navigation velocity for Dubins paths (m/s)
time_step 0.1 Simulation time step (s)
log_level info ROS log level (debug/info/warn/error/fatal)
record false Record a survey rosbag (see Recording a run)
bag_prefix boustrophedon Output bag directory prefix; a timestamp is appended
debug_gui false Start the rqt_console / rqt_graph GUIs. Leave false for headless / batch / parallel runs (they need a display); set true while debugging interactively

Survey geometry (edit in boustrophedon_run.py)

These are members at the top of __init__, not launch args. They must stay consistent with the sensor model and the NBV/search config for a fair comparison.

Member Default Notes
fov_x_deg 150.0 Horizontal (cross-track) FOV; must match the sensor model (compute_intrinsics.py)
fov_y_deg 25.0 Vertical (along-track) FOV
mount_angle_deg 20.0 Sensor down-tilt; must match the URDF fls_mount_angle_deg
clearance 5.0 Altitude above the seabed (m); survey_depth is derived from it
roi_north_min/max, roi_east_min/max 0 / 50 Survey rectangle (NED m); matches the 50x50 arena

Recording a run

Pass record:=true to capture a rosbag for offline analysis (CIR, coverage, detection timing):

ros2 launch baseline_mission baseline_mission_launch.py environment:=env_50x50_centroidBox record:=true

Each run is saved as its own directory under data/bags/, named <bag_prefix>_<timestamp>/. The timestamp (YYYYMMDD_HHMMSS) is always appended automatically, so runs never collide. bag_prefix is just a human-readable label on the front; it defaults to boustrophedon, so the command above produces e.g. data/bags/boustrophedon_20260617_151640/. Override it to tag the run's configuration:

ros2 launch baseline_mission baseline_mission_launch.py environment:=env_50x50_centroidBox record:=true bag_prefix:=boustrophedon_clearance5
# -> data/bags/boustrophedon_clearance5_20260617_151640/

Inside each run directory you get the .mcap file plus a metadata.yaml (written last, on clean close — its presence means the bag finalized correctly).

  • Topics: /face_hits (CIR (geometry_id, primitive_id) pairs), /detected_boxes (box centroids, published only when something is detected), /tf, /tf_static.
  • Format: MCAP — self-describing (the custom FaceHits decodes in Python without sourcing ROS) and crash-robust. Requires sudo apt install ros-humble-rosbag2-storage-mcap.
  • Location: <workspace-root>/data/bags/<bag_prefix>_<timestamp>/ — run ros2 launch from the workspace root. data/bags/ is gitignored; keeper bags are uploaded to external storage where needed.
  • Auto-stop: when the survey finishes, boustrophedon_run exits; an OnProcessExit handler shuts the launch down, which SIGINTs every process and cleanly finalizes the bag. So a recorded run is hands-off (no Ctrl-C needed).

Inspect a finished bag:

ros2 bag info data/bags/<bag_prefix>_<timestamp>

This is the same topic set, format, and layout as the NBV mission launch (demo_behaviors), so baseline and NBV runs are directly comparable. The boustrophedon survey duration is the natural time budget to give the NBV+search mission for a time-bounded comparison.

Related Packages

  • vista_sim — simulator, Dubins pose_to_pose action server, sensor model
  • demo_behaviors — the NBV mission this baseline is compared against
  • uuv_interfaces — PoseToPose.action, FaceHits.msg