Bowling Target Navigation — RZ/V2N Robot

Autonomous bottle detection and navigation for the Renesas RZ/V2N mecanum robot. DRP-AI hardware-accelerated YOLO detection, Cartographer SLAM, holonomic VFH navigation, all in a fullscreen GTK3 GUI.

Architecture

3 processes on 3 CPU cores:

Core	Process	Modules	Rate
0	GUI (main)	GTK3 rendering, settings, user input	30fps
1	ROS2 + Navigation	/map, /scan, TF2, control loop, Navigator	20Hz
2	Camera + DRP-AI	C++ stream inference, distance estimation	~15fps

IPC via SharedMemory (frames/map/laser), mp.Array (pose/nav state), mp.Queue (commands).

Set BOWLING_NAV_MULTIPROCESS=0 for single-process threading mode.

Detection Pipeline

DRP-AI C++ Binary (stream mode)
    │
    ├── Camera frames → /dev/shm/v2n_camera (zero-copy)
    └── JSON detections → stdout
            │
            ▼
    camera_worker.py filters:
    1. Confidence threshold (default 0.30)
    2. Min box size (10x15px — reject noise)
    3. Max box size (80% frame — reject false positives)
    4. Aspect ratio (bottle taller than wide)
    5. Bbox size tracker (rolling median, reject outliers)
            │
            ▼
    Distance estimator (bbox height → meters)
            │
            ▼
    IPC → ROS process → Navigator

Navigation Engine

navigate_to_target(target)
    │
    ├── Distance fusion: camera (primary) + LiDAR (fallback when bbox clipped)
    ├── Sensor offset compensation (camera 14cm, LiDAR 5cm from center)
    │
    ├── Far (>0.40m): Full VFH navigation with obstacle avoidance
    │   └── Speed scales with distance, mecanum strafing through gaps
    │
    ├── Close (<0.40m): Direct approach at min_speed (0.07 m/s)
    │   └── Bypasses VFH — bottle IS the obstacle
    │
    ├── Arrival (≤0.10m): Multi-signal check (vision + LiDAR + bbox_clipped)
    │   └── 0.3s temporal confirmation
    │
    └── Target lost:
        ├── <3s: Drift forward at min_speed
        ├── Close + lost: Blind approach (dead-reckon via TF2 map-frame)
        └── >3s: Search (360° scan → Archimedean spiral)

Settings (GUI)

Only 2 detection sliders (everything else has safe hardcoded defaults):

Setting	Range	Default	Description
Confidence	0.05–0.95	0.30	Lower = more detections, more false positives
Detection Memory	0.3–5.0s	1.0s	How long to keep a lost detection

Navigation params tunable in Nav/Search/Blind/Obstacle tabs. Sensor mounting in Setup tab. All auto-saved to ~/.config/bowling_target_nav/calibration.json.

Key Files

File	Purpose
nodes/main_gui.py	Entry point — spawns 3 processes or 3 threads
threads/ros_node.py	ROS2 node + 20Hz control loop
threads/camera_worker.py	DRP-AI detection + filtering
nav/navigator.py	Holonomic nav, VFH, blind approach, spiral search
nav/target_selector.py	Pick closest valid detection
state/detection_store.py	All tunable params + calibration persistence
gui/settings_window.py	8-tab settings UI
gui/panels/map_panel.py	SLAM map + diagnostics rendering
gui/panels/camera_panel.py	Camera feed + detection overlay
ipc/hub.py	SharedMemory IPC hub
processes/ros_process.py	ROS process with inbound/outbound sync
processes/camera_process.py	Camera process with settings watcher
detectors/drp_binary_detector.py	DRP-AI stream + pipe backends

Deploy to V2N

# From PC (use /* glob, NOT trailing /)
scp -r src/bowling_target_nav/bowling_target_nav/* \
  root@192.168.50.1:/root/ros2_ws/src/bowling_target_nav/bowling_target_nav/

# Clear Python cache on V2N
ssh root@192.168.50.1 "find /root/ros2_ws -name '__pycache__' -exec rm -rf {} + 2>/dev/null"

# Restart GUI
ssh root@192.168.50.1 "pkill -f main_gui; pkill -f bringup"
# Then press the floating launcher button on screen, or:
ssh root@192.168.50.1 "cd /root/ros2_ws/src/bowling_target_nav/scripts && ./bowling_gui.sh &"

TF Tree

map ──► odom ──► base_link ──► laser
  │       │           ├──► camera_link
  │       │           └──► wheels
  │       └── odometry_node (encoder ticks)
  └── cartographer (SLAM)

Critical: Cartographer must use provide_odom_frame=false + published_frame="odom" since odometry_node already publishes odom→base_link. Constraint builder disabled on V2N aarch64 to prevent mutex deadlock.

ROS2 Topics

Topic	Type	Direction	Purpose
/cmd_vel	Twist	Publish	Motor velocity
/scan	LaserScan	Subscribe	LiDAR points
/map	OccupancyGrid	Subscribe	SLAM grid
/odom	Odometry	Subscribe (via TF)	Robot pose
/gui/command	String	Subscribe	Remote GO/STOP
/arduino/cmd	String	Publish	Motor calibration

License

MIT