3. High-Level System Architecture

This autonomous mobile robot system operates in two phases: (A) Pre-Mapping to build a static 2D occupancy grid via teleoperated SLAM, and (B) Mission to autonomously follow a moving ArUco marker-bearing target while avoiding unknown obstacles. The architecture follows the Perception → Estimation → Planning → Actuation convention.

3.0 Reference links (libraries + key interfaces)

3.1 System Data Flow Diagrams

3.1.1 Mission Mode Data Flow (Perception → Estimation → Planning → Actuation) 

graph LR
    subgraph Perception["🔹 PERCEPTION"]
        Lidar["LiDAR Driver<br/>/scan"]
        Camera["Camera Driver<br/>/camera/image_raw"]
        ArUco["ArUco Detector<br/>/detected_aruco_pose<br/>(Custom)"]
    end

    subgraph Estimation["🔹 ESTIMATION"]
        AMCL["AMCL<br/>/amcl_pose<br/>(Library)"]
        Costmap["Local Costmap<br/>obstacle layer<br/>(Library)"]
        TF["TF2 System<br/>frame transforms<br/>(Library)"]
        GoalGen["Goal Generator<br/>/target_goal_pose<br/>(Custom)"]
        Coord["Target Coordinator<br/>state machine<br/>(Custom)"]
    end

    subgraph Planning["🔹 PLANNING & CONTROL"]
        Nav2["Nav2 Stack<br/>planner + controller<br/>(Library)"]
    end

    subgraph Actuation["🔹 ACTUATION"]
        Motor["Base Controller<br/>motor commands<br/>(Library)"]
    end

    Lidar --> AMCL
    Lidar --> Costmap
    Camera --> ArUco
    ArUco --> GoalGen
    ArUco --> Coord
    AMCL --> TF
    TF --> GoalGen
    GoalGen --> Coord
    Coord --> Nav2
    Costmap --> Nav2
    Nav2 --> Motor

    style Perception fill:#a8e6cf
    style Estimation fill:#ffe66d
    style Planning fill:#ff6b6b
    style Actuation fill:#4ecdc4

3.1.2 State Machine: Target Follow + Loss Recovery 

stateDiagram-v2
    [*] --> FOLLOW: Mission start / target visible
    FOLLOW --> LOST: 5+ frame misses AND LOS blocked
    LOST --> FOLLOW: ArUco re-detected
    LOST --> SEARCH: Timeout expires
    SEARCH --> FOLLOW: ArUco detected
    SEARCH --> SEARCH: Rotating/scanning

    note right of FOLLOW
        Dynamic goal updates.
        Nav2 tracking path.
    end note

    note right of LOST
        Goal frozen at last pose.
        Nav2 replans around obstacles.
    end note

    note right of SEARCH
        Rotational sweep for
        target re-acquisition.
    end note

3.2 Module Declaration Table

Module Name Type ROS 2 Package Inputs Outputs
LiDAR Driver Library Driver-specific Hardware /scan (sensor_msgs/LaserScan)
Wheel Odometry Library Base firmware Motor encoders /odom, /tf (odom→base_link)
Camera Driver Library usb_cam / cv_camera Hardware /camera/image_raw (sensor_msgs/Image)
SLAM Toolbox Library slam_toolbox /scan, /tf, /odom /map, /tf (map↔odom)
Map Server Library nav2_map_server Disk file /map (nav_msgs/OccupancyGrid)
AMCL Library nav2_amcl /scan, /map, /tf /amcl_pose, /tf (map→odom)
Local Costmap Library nav2_costmap_2d /scan, /tf, /amcl_pose Costmap layers (obstacle)
Nav2 Planner Library nav2_navfn_planner /map, /costmap, /tf /plan (nav_msgs/Path)
Nav2 Controller Library nav2_dwb_controller /costmap, /tf, /path /cmd_vel (geometry_msgs/Twist)
ArUco Detector Custom Custom (this project) /camera/image_raw /detected_aruco_pose (geometry_msgs/PoseStamped)
Goal Generator Custom Custom (this project) /detected_aruco_pose, /tf /target_goal_pose (geometry_msgs/PoseStamped)
Target Coordinator Custom Custom (this project) /detected_aruco_pose, /amcl_pose, /scan, costmap /navigate_to_pose action goals
TF2 Broadcaster Library tf2_ros Sensor poses, odometry /tf, /tf_static
Base Controller Library Base firmware /cmd_vel Motor PWM commands

3.3 Module Intent Writeups

Library Modules

  • LiDAR Driver: Publishes 2D laser scans at 10–25 Hz. Tune: scan rate, max range, angular resolution (typical: 25 Hz, 8 m range).

  • Wheel Odometry & Base Controller: Encoder-based odometry at ~50 Hz; controller translates /cmd_vel to motor PWM. Tune: wheel radius, track width, drift compensation (<30 cm over 30 sec).

  • Camera Driver: Publishes RGB frames at 15–30 Hz with precomputed intrinsic calibration. Tune: resolution, frame rate, distortion coefficients.

  • SLAM Toolbox: Standard ROS 2 SLAM front-end for 2D LiDAR. Builds occupancy grid via scan matching and loop closure. Tune: loop closure threshold, min travel distance, occupancy threshold.

  • Map Server: Loads precomputed occupancy grid from disk for mission phase.

  • AMCL: Localizes against static map by matching LiDAR scans; corrects odometry drift. Tune: particle count (50–500), scan matching model, update rates.

  • Local Costmap: Maintains rolling 5×5 m window, marks pre-mapped and dynamic obstacles. Tune: decay rate, inflation radius, update frequency.

  • Nav2 Planner & Controller: Global planner (NavFn) + local controller (DWB). Tune planner: potential scale, lethal cost threshold. Tune controller: max velocity, acceleration limits, lookahead distance.

  • TF2 System: Manages coordinate frame transformations (map ← odom ← base_link). Configure: static transforms (sensor offsets), dynamic transforms (AMCL, odometry).

  • Base Controller: Translates motor commands from Nav2.

Custom Modules

ArUco Detector Implements real-time visual marker detection using OpenCV. Algorithm: (1) subscribe to /camera/image_raw at camera frame rate; (2) apply adaptive thresholding to detect candidate marker corners; (3) validate corner patterns against ArUco dictionary (DICT_5X5_100); (4) reject false positives using confidence threshold; (5) solve PnP with known marker size and camera intrinsics to compute 3D pose in camera frame; (6) publish geometry_msgs/PoseStamped to /detected_aruco_pose. Success criteria: detects target marker at 0.5–3 m with <5% pose error; tolerates 1–2 frame intermittent occlusions; rejects false positives >95%.

Goal Generator from ArUco Transforms target pose from camera frame to map frame and computes reachable goal. Algorithm: (1) subscribe to /detected_aruco_pose; (2) query TF2 for camera→base_link→odom→map chain; (3) transform pose to map frame; (4) offset goal 0.5 m behind target (approach direction); (5) apply moving-average temporal smoothing over last N detections; (6) publish /target_goal_pose at ~10 Hz. Ensures goal is always reachable by the robot. Success criteria: <0.1 m transform error; goal always reachable; temporal jitter <0.2 m s.d.

Target Follow + Loss Recovery Coordinator Implements state machine (FOLLOW → LOST → SEARCH) to handle target occlusion and recovery. Algorithm: FOLLOW state: continuously query /detected_aruco_pose and Goal Generator’s /target_goal_pose; send goals to Nav2 at 5–10 Hz. Transition to LOST: detect ≥5 consecutive frame misses AND costmap raytrace confirms line-of-sight blockage; freeze goal at last-known target; start loss timer (2–5 sec). LOST state: goal is static; monitor for re-detection or clearing. Transition to SEARCH: loss timeout expires and ArUco still absent. SEARCH state: execute rotational sweep or frontier-based movement for 15–30 sec; on re-detection, immediately transition to FOLLOW. Success criteria: detect occlusion within 0.5 sec; re-acquire target within 10 sec; recover 90%+ of loss events.

3.4 Terminology Reference

Term Definition
Pre-mapping Phase 1: teleoperated SLAM exploration to build offline 2D occupancy grid (see slam_toolbox).
Mission Phase 2: autonomous target-following with localization + navigation (see Nav2).
Line of Sight (LOS) Direct unobstructed visual path from camera to ArUco marker (see OpenCV ArUco).
Occlusion-induced loss Target loss caused by obstacle blocking LOS.
Unknown obstacle Real-time obstacle not in pre-mapped grid; detected by LiDAR (see LaserScan).
Local costmap Rolling window costmap updated from sensors; used by Nav2 controller (see Costmap 2D).
Recovery behavior Fallback strategy (e.g., search rotation) for target re-acquisition (Nav2 behavior tree / recovery patterns; see Nav2 docs).
This site uses Just the Docs, a documentation theme for Jekyll.