Skip to main content

🛠️ Customization & Upgrades

The LIMO Pro + MyCobot + YOLO platform is not just plug-and-play — it’s designed to be modified, extended, and upgraded.
This page is your hands-on guide for changing how the robot works, adding new capabilities, and pushing it beyond its default setup.

🗺 Mapping & Exploration​

1. Switch from Cartographer to RTAB-Map​

  • Why: RTAB-Map enables real-time autonomous exploration without pre-building maps.
  • How:
    1. Install rtabmap_ros: 
      sudo apt install ros-foxy-rtabmap-ros
    2. Replace Cartographer launch in full_system.launch.py with rtabmap.launch.py.
    3. Use rtabmap_explore to roam unmapped spaces automatically.
  • Hint: RTAB-Map is heavier on resources — keep other processes light for smoother performance.

2. Tune SLAM Parameters​

  • Why: Better loop closure, less drift, higher mapping accuracy.
  • How:
    Edit parameters in cartographer.lua or rtabmap_params.yaml:
    • Increase loop closure frequency
    • Adjust voxel size for your space
  • Hint: Test changes in a small area before applying to full missions.

🎯 Object Detection & Recognition​

3. Improve YOLO Accuracy​

  • Why: Current model trained with only ~150 images; accuracy can reach 99%+ with a richer dataset.
  • How:
    • Use Roboflow to expand dataset & annotate more images.
    • Fine-tune via a custom train_yolo.py script in ROS2.
    • Replace the .pt file in object_detector.py.
  • Hint: Capture images in the same lighting and environment the robot will operate in.

4. Add More Object Classes​

  • Why: Allow robot to identify multiple object types and act differently for each.
  • How:
    • Add new labeled classes to the YOLO dataset.
    • Retrain, then modify object_detector.py to filter cls_name.
  • Hint: Use a class-to-action mapping table in Python to control the arm’s behavior per object.

5. Enable Configurable Classes at Runtime​

  • Why: Change detection targets without retraining.
  • How:
    Pass --classes to YOLO predict or use a ROS2 parameter for active detection classes.
  • Hint: Add a dynamic parameter server node so you can change classes mid-mission.

🧠 Memory & AI Enhancements​

6. Add Vector Memory for User Recognition​

  • Why: Robot learns the user’s face and returns thrown objects only to them.
  • How:
    • Integrate face_recognition or dlib with a local embedding database.
    • Link recognition results to mission logic in mission_manager.py.
  • Hint: Store embeddings in JSON/SQLite for quick retrieval; keep images under consistent lighting.

7. AI Path Planning​

  • Why: Optimize pickup order when multiple objects are detected.
  • How:
    • Use A* or Dijkstra with dynamically updated object positions.
    • Or integrate stable-baselines3 RL agent to learn optimal routes.
  • Hint: Keep a fallback static waypoint list in case AI planner fails.

🤝 Multi-Robot Coordination​

8. Collaborative Missions​

  • Why: Multiple LIMOs share the workload, mapping, and object collection.
  • How:
    • Share maps via map_server and DDS multicast.
    • Create a master mission manager node to divide waypoints.
  • Hint: Assign each robot a namespace (/robot1, /robot2) to avoid topic collisions.

🖥 UI & User Interaction​

9. Browser-Based Dashboard​

  • Why: Control and monitor the robot remotely.
  • How:
    • Install rosbridge_server + web_video_server.
    • Build a web app with roslibjs or integrate Foxglove Studio.
  • Hint: Expose only safe commands in remote mode.

10. Voice-Enabled Conversation​

  • Why: Control robot hands-free and add interactive personality.
  • How:
    • Use vosk or whisper.cpp for offline speech-to-text.
    • Map recognized commands to ROS2 actions.
  • Hint: Add audio feedback so the user knows the robot understood the command.

⚡ Performance Optimizations​

11. Nav2 & Motion Tuning​

  • Why: Smoother navigation, fewer localization errors.
  • How:
    Adjust velocity/acceleration in nav2_params.yaml:
    • Lower max_vel_theta for precise turning.
    • Fine-tune planner_frequency for better responsiveness.
  • Hint: Record bag files before/after tuning to measure improvements.

12. Offload Heavy Processing​

  • Why: Reduce CPU load for smoother multi-node operation.
  • How:
    • Run YOLO on an edge GPU (e.g., Jetson).
    • Use nodelets or multi-threaded executors.
  • Hint: Cache TF transforms in object_detector.py to cut processing time.
Pro Tip

Upgrades don’t have to be huge — even small tweaks (like better SLAM tuning) can boost mission success rates dramatically.

on August 8, 2025 by KRISH RAY