Target Interception Simulation

ROS 2 simulation sandbox for moving object tracking and interception using an Ackermann-steered vehicle. We incrementally increase the complexity of the scenario (v0, v1, v2, ...) to validate simpler logic and theory before moving onto more complexity.

An Ackermann vehicle has a constant forward velocity, with steering is limited by the 0.24m wheelbase and 0.6rad steering limit; linear acceleration and deceleration are also bounded. We assume the target is initially visible.

Architecture

All versions have the same RGB-D perception and target-tracking pipeline. What changes across v0, v1, and v2 is the ego-pose source, command structure, and world complexity.

Package	Description
robot_description	Ackermann vehicle geometry and static sensor extrinsics
robot_sim	Gazebo arena, sensors, bridges, control, and trial evaluation
robot_interfaces	Stamped 2D target-detection message
robot_perception	HSV detection and synchronized RGB-D projection
robot_tracking	Constant-velocity target Kalman filter
robot_navigation	Intercept solve and direct Ackermann pursuit control
robot_odometry	Wheel/RGB-D ego odometry fusion and initial map alignment

v0. Open Space

Data flow

flowchart LR
  camera["RGB-D camera"] --> target["Target perception + tracking"]
  ego["Ground-truth ego pose"] --> target
  target --> control["Direct intercept control"]
  ego --> control
  control --> vehicle["Vehicle"]

Loading

Scenario

The vehicle directly pursues a red ball moving at 0.4 m/s on a 3 m circle in an empty 12 × 12 m arena. Target position comes from RGB-D perception, while ego pose comes from ground truth. The constant-velocity target filter is intentionally model-mismatched with the circular motion.

v1. Obstacles

Data flow

flowchart LR
  camera["RGB-D camera"] --> target["Target perception + tracking"]
  ego["Ground-truth ego pose"] --> target
  target --> supervisor["Interception supervisor"]
  ego --> supervisor
  supervisor --> control["Nav2 + terminal pursuit"]
  control --> vehicle["Vehicle"]

Loading

Scenario

The same moving target is placed in an arena with two fixed chicane barriers. Nav2 routes toward predicted intercept goals using the known static map, then hands control to direct terminal pursuit near the target. Ego pose remains ground truth; obstacles are map-known and are not sensed dynamically.

v2. Localization

Data flow

flowchart LR
  camera["RGB-D camera"] --> target["Target perception + tracking"]
  camera --> localization["Visual + wheel localization"]
  wheel["Wheel odometry"] --> localization
  target --> interception["v1 interception stack"]
  localization --> interception
  initial["Known initial pose"] --> interception
  interception --> vehicle["Vehicle"]

Loading

Scenario

v2 reuses the v1 arena, target motion, and navigation strategy, but replaces application use of ego truth with fused wheel and RGB-D odometry. A launch-time map -> odom transform seeds the known initial pose; ground-truth robot and target odometry are reserved for trial evaluation.

Build and Run

In ROS 2 Jazzy:

rosdep install --from-paths src --ignore-src -r -y
colcon build --symlink-install
source install/setup.bash

# Run scenarios
ros2 launch robot_sim intercept.launch.py
ros2 launch robot_sim v1_intercept.launch.py
ros2 launch robot_sim v2_intercept.launch.py

Results

We run ten trials scenarios that vary the target phase and vehicle pose while keeping the target initially visible. Results include capture and contact counts, clearances, estimation errors, and termination reasons.

• v0: Target captured in all 10 scenarios. Capture times ranged from 4.47s to 7.99s, with target-position RMSE between 0.014m and 0.016m.
• v1: Target captured in all 10 scenarios with no reported fixed-obstacle contacts.
• v2: Target captured in 8 of 10 scenarios with no observed contacts. One trial timed out and one exceeded the process timeout, leaving localization data incomplete.

Across completed trials, position RMSE ranged from 0.158 m to 0.403 m, yaw RMSE from 0.072 rad to 0.324 rad, final position error from 0.072 m to 0.650 m, and localization availability from 91.8% to 99.5%. These results do not consistently meet the targets of 0.20 m position RMSE, 0.15 rad yaw RMSE, 0.30 m final position error, and 95% availability.

Roadmap

Version	Increment
v0	Open-space interception using ego ground truth
v1	Obstacles and Nav2 for mid-course routing, with direct terminal pursuit
v2	Fused wheel/RGB-D ego odometry with known initial map alignment
v3	Search, loss recovery, reset handling, and explicit mission states