By Or Tslil, Amit Elbaz

Paper link

ROS implementation of online semantic SLAM, based on the a published paper - "Representing and updating object identities in semantic SLAM". The object detection node is based on SSD300 architecture and forked from https://github.com/balancap/SSD-Tensorflow.

The following python packges are required:

• python 2.*
• numpy
• sklearn
• sciPy
• openCV
• TensorFlow 1.1* (GPU version)
• hector_mapping (http://wiki.ros.org/hector_mapping)
• currently tested in ros melodic in ubuntu 18.04

1. Download repository to your catkin workspace:

git clone https://github.com/or-tal-robotics/object_map.git

2. Build:

catkin_make

3. Install SSD image detector for ROS:

pip install -e object_detector_ssd_tf_ros

4. Unzip SSD weights in object_map/object_detector_ssd_tf_ros/ssd/model/ssd_300_vgg.ckpt.zip

• For a demo simulation use:

roslaunch gazebo_demo demo.launch

• For a demo simulation working with the "Bhattacharyya coefficient" method of updating the map use:

roslaunch gazebo_demo Test.launch