prox_mpc_obstacle_tracker

An in-house 2D-lidar dynamic-obstacle detector and tracker for ProxMPC.

A managed lifecycle node clusters a sensor_msgs/LaserScan, transforms the cluster centroids into a fixed tracking frame, runs one IMM (constant-velocity + constant-turn-rate) filter per object, and publishes the confirmed tracks - including sampled predicted positions along each track’s estimated arc - as a prox_mpc_msgs/ObstacleArray. That feed is what prox_mpc_controller consumes for predictive (dynamic) obstacle avoidance.

The detection and tracking math is written from scratch (Eigen only, no third-party tracker), so the package is license-clean and unit-testable without ROS. The design, algorithm, parameters, and interfaces are documented in doc/architecture.md.

Table of Contents

• Key Features
• Prerequisites
• Build
• Run
• Interfaces
• Lifecycle
• Composition
• Testing
• License

Key Features

• Lifecycle node: managed configure -> activate -> deactivate -> cleanup, with a signal-safe shutdown ladder in the standalone driver.
• Self-contained pipeline: LaserScan -> planar points -> adjacency clusters -> tracking-frame centroids -> IMM (CV + CTRV) tracks with sampled predicted positions.
• Multi-object tracking: gated greedy nearest-neighbour association, one IMM filter per track (a constant-velocity Kalman filter and a constant-turn-rate-and-velocity EKF run in parallel, blended by model probability), and a birth/confirm/death lifecycle. imm_enabled: false restores the legacy single-CV path.
• Curved prediction feed: each published obstacle carries prediction_steps predicted positions at prediction_dt spacing (default 25 x 0.1 s = 2.5 s), so the controller can follow turning obstacles instead of a straight constant-velocity ray.
• Wall rejection: a cluster-radius cap drops extended structure (walls) whose centroid would otherwise be tracked as a phantom fast-moving obstacle.
• Pure core: the clustering and tracker are a ROS-free Eigen library (prox_mpc_obstacle_tracker_core) covered by GoogleTest.

Prerequisites

• ROS 2 Jazzy on Ubuntu 24.04.
• prox_mpc_msgs (workspace package).
• Eigen3, rclcpp, rclcpp_components, rclcpp_lifecycle, lifecycle_msgs, sensor_msgs, geometry_msgs, tf2, tf2_ros (resolved by rosdep).

Build

colcon build --symlink-install --packages-select prox_mpc_msgs prox_mpc_obstacle_tracker
source install/setup.bash

Run

The standalone executable is a self-activating lifecycle node: it brings itself up (configure -> activate), spins, and tears itself down on SIGINT/SIGTERM. The bundled launch file loads config/obstacle_tracker.yaml and wires the node-only logger level, with a params_file argument to override the parameters:

ros2 launch prox_mpc_obstacle_tracker obstacle_tracker.launch.py
ros2 launch prox_mpc_obstacle_tracker obstacle_tracker.launch.py \
  params_file:=/path/to/custom.yaml

To run the executable directly instead of through the launch file:

ros2 run prox_mpc_obstacle_tracker obstacle_tracker \
  --ros-args --params-file \
  $(ros2 pkg prefix prox_mpc_obstacle_tracker)/share/prox_mpc_obstacle_tracker/config/obstacle_tracker.yaml

config/obstacle_tracker.yaml is the single source of truth for the parameters. Watch the output with:

ros2 topic echo /tracked_obstacles

In simulation the tracker is started automatically by the demo’s Nav2 launch with predictive:=True (see prox_mpc_demo/doc/nav2-simulation.md).

Interfaces

File truncated at 100 lines see the full file

Changelog for package prox_mpc_obstacle_tracker

1.0.0 (2026-07-28)

• Initial release: lifecycle node that clusters a 2D LaserScan, associates clusters to an IMM (constant-velocity + constant-turn-rate) filter in a fixed frame, and publishes a prox_mpc_msgs/ObstacleArray.
• Composable component registration plus a mutually-exclusive callback group and teardown guard, so it is safe under a MultiThreadedExecutor; node-only log_level parameter and a standalone launch file.
• Contributors: Simone Contorno