Robotics-Final-Project

The final project of the Robotics course. A robot has to find it's way out of a maze with the VFH algorithm. This robot should also detect lanes and traffic signs in a city setting and act accordingly.

Requirements:

• ROS Noetic
• turtlebot3, turtlebot3_msgs and turtlebot3_simulations, turtlebot3_autorace_2020
• numpy
• ultralytics
• opencv-python

How do I build this?

1. Install the requirements.
2. Clone this package beside your turtlebot3 packages:

foo@bar:~/catkin/src$ git clone https://github.com/rzninvo/robotics_final_project.git

3. Navigate to the root directory of your catkin workspace.
4. Do catkin_make:

foo@bar:~/catkin$ catkin_make

First Phase: The VFH Pathfinding Algorithm

In this phase, our goal is to implement pathfinding and obstacle avoidance using the Vector Field Histogram (VFH) algorithm. The robot (which is a turtlebot3 robot) should start at the starting position of $(0, 0)$ in the world of updated_maze.world located in the /worlds folder. The objective of this robot is to avoid all the obstacles in the maze and find it's way to the end of the maze which is at the goal position of $(13, 7)$.
This package implements the VFH Algorithm based on this paper.

Our changes to the VFH Algorithm:

In this package, some changes were done by us to further optimize the algorithm. Here is a brief explanation:

1. Changes to the Confidence Calculation:
   Since in our implementation, the Lidar Scanner only gives us the distance to any object surounding the robot and not it's confidence of being an actual object, we implemented our own confidence formula:
   $c_\theta = |\frac{d}{ws} - 1|$
   Where $c_\theta$ is the confidence of the corresponding laser reading angle, $d$ is the laser reading distance, and $ws$ is our window size.
   This formula makes it so that the nearest object to the robot is close to $1$ and the farthest object in the Active Window is close to $0.5$. This formula has proven to have a better Polar Density Histogram than setting it to the default confidence of $1$ for every reading in the Active Window.
2. Filtering out certain angles:
   In some situations where there is no possible Candidate Valley other than the angles between $(135\degree, 225\degree)$, the robot would choose these valleys as it's nearest valley to the target and attempt to make a $180\degree$ turn. To avoid these situations, we have filtered out the aforementioned angles and don't recognize them as candidate valleys.

Components

1. The VFH Algorithm Node:
   This node acts as a service node, plotting the Polar Density Histogram of the robot and constantly giving our robot it's best rotating angle to the goal direction while avoiding obstacles using the VFH algorithm. It's message type is found here.
2. The VFH Pathfinder Node:
   This node acts as the control node, while having a proxy to the vfh_planner_service and feeding on it's calculated output angle and $h_c$. This calculated output angle acts as the error of a PID control for the angular velocity of our robot. The output $h_c$ acts as the control for the robot's linear velocity based on the VFH paper.

How do I run this?

1. Navigate to the root directory of your catkin workspace.
2. Source your workspace:

foo@bar:~/catkin$ . devel/setup.bash

3. Launch the provided launch file:

foo@bar:~/catkin$ roslaunch robotics_final_project vfh_planning.launch

Expected Output Video

1688666725140.mp4

Second Phase: Autorace Lane Detection and Traffic Sign Detection

In this phase, we have implemented a robot which can detect roads, follow the road lane, detect traffic signs, and act according to the traffic sign.

First Part: Lane Detection

Based on the documentation provided in this link, we have used the detect_lane node to detect the road line and move alongside it. However there are obstacles in the robot's road path and the robot should avoid hitting these obstacles.
To achieve this, we have edited the control_lane node located at turtlebot3_autorace_2020/turtlebot3_autorace_detect/nodes/control_lane and have made our 'control_lane' version of this node.

How do I run this?

1. Navigate to the root directory of your catkin workspace.
2. Source your workspace:

foo@bar:~/catkin$ . devel/setup.bash

3. Launch the provided launch file:

foo@bar:~/catkin$ roslaunch robotics_final_project lane_detection.launch

Note

In order to launch this section correctly, you should replace your detect_node file located at the turtlebot3_autorace_2020/turtlebot3_autorace_detect/nodes/detect_lane path with the provided detect_node file.

Second Part: Traffic Sign Detection

Now that the robot can follow the road line, it should also be able to detect the traffic signs and act upon them. The documentation provided in this link uses the SIFT Algorithm to detect traffic signs. We have tweaked some of the detection nodes provided by this package and have came up with the four detection nodes located at here. After tweaking the nodes to publish to our desired topics, we also tweaked the existing control_lane in our package to act upon any detection of traffic signs.

How do I run this?

1. Navigate to the root directory of your catkin workspace.
2. Source your workspace:

foo@bar:~/catkin$ . devel/setup.bash

3. Launch the provided launch file:

foo@bar:~/catkin$ roslaunch robotics_final_project traffic_sign_detection.launch

Note

In order to launch this section correctly, you should replace your detect_node file located at the turtlebot3_autorace_2020/turtlebot3_autorace_detect/nodes/detect_lane path with the provided detect_node file.

Expected Output Video

Traffic_Sign_Detection_Lane_Detection.mp4

Third Phase: YOLO Sign Detection

We used the SWIFT algorithm to detect traffic signs, but we also wanted to use a different approach. That's why we used the YOLOv8 Object Detecton model. We decided to train the yolov8n.pt model using the Traffic and Road Signs dataset. The training results can be seen in the /yolo directory.

Components

1. The Yolov8 Sign Detector Node:
   This node does all the traffic sign detections. It subcribes to the camera image input and uses the trained YOLO model to annotate and predict. If our model detects a traffic sign which it recognizes, it will publish a signal to it's corresponding topic.
2. The YOLO Control Lane:
   This node acts just like the previous control lane nodes, but the topics to which it subcribes and acts upon are different.

How do I run this?

1. Navigate to the root directory of your catkin workspace.
2. Source your workspace:

foo@bar:~/catkin$ . devel/setup.bash

3. Launch the provided launch file:

foo@bar:~/catkin$ roslaunch robotics_final_project yolo_sign_detection.launch

Note

To be able to run this launch file correctly, you should copy the model folders inside the /models directory to the turtlebot3_simulations/turtlebot3_gazebo/models directory.

Expected Output Video

YOLO_Sign_Detection.mp4