environment_mapper

Environment Mapper is a ROS2-based toolkit that integrates camera/IMU input, NVIDIA DeepStream object detection, and RTAB-Map semantic mapping. It provides launch files for sensors and mapping tools, plus an object_detection package that uses DeepStream to detect and track objects, then labels them in RTAB-Map maps.

Repository Contents

• `` — ROS2 launch scripts and detailed build notes for CUDA, DeepStream, RTAB-Map, and RealSense.
• `` — A ROS2 ament package containing:
  • DeepStream GStreamer pipeline (detection_pipeline.py)
  • ROS2 nodes: object_detector, object_labeler, and rgbd_merger
  • Pre-trained models/configs under Primary_Detector/
• `` — Utilities like a Kalibr Dockerfile.

Data Flow

1. Camera node publishes RGB and depth.
2. object_detector subscribes, converts frames via cv_bridge, and pushes them to DeepStream (appsrc).
3. DeepStream runs detection (nvinfer), tracking (nvtracker), and annotation (nvdsosd), then Python probes extract metadata with pyds.
4. Depth is averaged per detection ROI (if available).
5. Detections are published to /detections, consumed by object_labeler, which adds labels to RTAB-Map and publishes PoseStamped goals.

Quickstart

1. Install ROS2 Humble, NVIDIA drivers + CUDA, DeepStream SDK + Python bindings, librealsense2, and RTAB-Map.
2. Clone into a ROS2 workspace:

mkdir -p ~/ros2_ws/src
cd ~/ros2_ws/src
git clone https://github.com/toutia/environment_mapper.git
cd ~/ros2_ws
colcon build --symlink-install
source install/setup.bash

3. Run your chosen launch files for sensors and RTAB-Map, then start:

ros2 run object_detection object_detector
ros2 run object_detection object_labeler

Requirements & Caveats

• Requires NVIDIA GPU + DeepStream SDK.
• Model engine files may be platform-specific; rebuild if needed.
• ROS2-focused; some legacy ROS1 references exist.

Suggested Improvements

• Add requirements.txt and Dockerfile for reproducible builds.
• Provide lightweight example bag files for testing without hardware.
• Document exact topic names/parameters for all nodes.

License

• object_detection is under Apache-2.0; DeepStream configs/models follow NVIDIA licensing.