Object Localization and Relative Navigation on Gazebo

An object detection and relative navigation scheme for a custom robot on Gazebo. The project was developed as senior design project on Fall/21 semester at Bogazici University.

Report and Presentation

Project report and presentation can be accessed under resources directory under project tree.

Getting Started

Prerequisites

ROS Melodic

Install following the instructions on http://wiki.ros.org/melodic/Installation.

ROS Package Dependencies

Install them via,

sudo apt-get install ros-melodic-costmap-2d ros-melodic-robot-localization ros-melodic-yocs-cmd-vel-mux ros-melodic-effort-controllers ros-melodic-navigation ros-melodic-geometry2 ros-melodic-nmea-msgs ros-melodic-bfl ros-melodic-arbotix-python

Tensorflow Object Detection API

Install following the tutorial under the link,

https://tensorflow-object-detection-api-tutorial.readthedocs.io/en/latest/install
Python Dependencies

Activate the Conda environment created in Step 3 and run,
```
conda install opencv-python
```

Installing

Robot and Environment

Create a new Catkin workspace,
```
mkdir -p ~/navi_ws/src
```

Clone the repository to the workspace,

cd ~/navi_ws/src
git clone https://github.com/mehmetalici/localize-and-navigate.git

Build the packages in the workspace,
```
cd ~/navi_ws
catkin_make
```

Source the workspace and the robot,

cd ~/navi_ws
source devel/setup.bash
source src/localize-and-navigate/segway_v3/segway_v3_config/std_configs/segway_config_RMP_220.bash

Test the robot and environment running,
```
roslaunch segway_gazebo segway_empty_world.launch
```
You should see our robot and world in the Gazebo simulation software.

Detection

Download a model

Activate your environment, and run,
```
cd ~/navi_ws/src/localize-and-navigate/obj_detection/src
python download_model.py 20200711 ssd_mobilenet_v2_320x320_coco17_tpu-8
```
to download the model with name and date ssd_mobilenet_v2_320x320_coco17_tpu-8 and 20200711, respectively.

You can change the model name and date according to the models in the following list, which involves all pre-trained models, https://github.com/tensorflow/models/blob/master/research/object_detection/g3doc/tf2_detection_zoo.md

Tracking

Download a specific version of OpenCV Bridge.

Apply,

cd ~
git clone https://github.com/mehmetalici/opencv-bridge-err-resolver.git
cd opencv-bridge-err-resolver
catkin clean -b
catkin build

Robot Control

Control of Head

A pan-tilt device rotates the head of the robot. The joints of Pan-Tilt are controlled by their PID controllers.

The head features a ZED RGM-D camera and it is possible to control the head by providing (x, y) coordinates of the target in the RGB image. To test it, first run,

roscd robot
rosrun rviz rviz -d rviz/rviz_cfg.rviz

and For example, to move to a position of 0.5 rads positive to the axis of rotation for both joints, run,

rostopic pub /pan_controller/command std_msgs/Float64 "data: 0.5" 
rostopic pub /tilt_controller/command std_msgs/Float64 "data: 0.5"

Control of Navigation

The robot can be navigated by commanding to RMP220.

For instance, to move the robot through a circle, run,

rostopic pub -r 10 /segway/teleop/cmd_vel geometry_msgs/Twist "linear:
  x: 10.0
  y: 0.0
  z: 0.0
angular:
  x: 0.0
  y: 0.0
  z: 1.0"

Control of Sensors

The robot is equipped with a Velodyne Puck LIDAR and a ZED RGB-D Camera. The sensors are always active and publishing under their respective topics.

Sensor information can be visualized using RViz. To do this, follow the instructions here.

Detection and Tracking

Activate your environment and source your workspace.

Run,

roscd obj_detection/src
source ~/opencv-bridge-err-resolver/install/setup.bash

To detect a person, run,
```
python detect_track.py person
```
To see a complete list of avaliable objects you can pass to the script, go to the directory in which you downloaded the model and open mscoco_label_map.pbtxt.
To visualize detection, run,
```
rosrun rviz rviz
```
Then select the topics /rgb_with_det or /rgb_with_det, which produces images with boxes of target object, on RViz GUI.

Navigation

Source the workspace and run,

roscd obj_detection/src
python navigate_goal.py

Run the APF algorithm,
```
roscd apfrl/src
python apf_rl_0707.py
```