Introducing "TagBot" – our innovative project where robotics meets fun and fitness. Unlike typical indoor robotics activities, Tagbot takes tech outdoors, encouraging kids to dive into STEM learning while staying active.
Our mission is to create robots that can play tag with kids. With Tagbot, children don't just learn science and tech; they move, strategize, and grow holistically.
- Yahboom ROSMASTER X3
- Jetson Nano 4GB main control board
- Astra Pro Plus RGB and Depth camera
- 2D LiDAR
- Ubuntu 18.04 LTS
- ROS1 Melodic
- Yahboom Tutorial Code
- Python environment
- Clone this repository to your robot.
git clone https://github.com/AidanNowa/EC545_TAGBOT.git
- Register the TagBot workspace
tagbot_wsand its packages to your ROS environment.
source EC545_TAGBOT/tagbot_ws/devel/setup.bash
- Run the TagBot.
roslaunch controller controller.launch
The controller.launch file automatically starts all the ROS nodes that are necessary to execute the entire TagBot system.
You can also test the color_detector package alone with the following command:
roslaunch color_detector color_detector.launch
We implemented two ROS packages to achieve the TagBot system: color_detector and controller. They depend on sensor packages provided by Yahboom, namely astrapro and rplidar.
The color_detector looks for the closest object with a specified color and returns its distance and angle with respect to the ego robot. The default target color is red. This node subscribes to two ROS topics: /camera/rgb/image_raw and /camera/depth/image_raw. First, it locates red contours in an RGB image. Small contours are ignored as a noise. Second, it extracts the depth of each contour in the corresponding depth image. Finally, it picks the contour with the closest distance and calculates the object angle with respect to the robot. The angle calculation is based on the camera's Field of View (FOV) specification. The distance and angle information is published as a ROS topic, /tagbot/target_position. It only publishes this information when objects are detected.
The object avoidance block guides the robot to turn away from detected objects or walls. This block depends on a ROS topic for 2D LiDAR measurements, /scan. The LiDAR has front field of view (FOV) of 320 degrees, with 40 degrees of blind spot on behind. For simplicity, we devided the 320 FOV into 3 directions: left, front, and right. For each direction, it flags a warning when there are more than a certain number of point clouds within a threshold distance (0.6 meters). When warnings are present, the robot stops and rotates on the spot to turn away from the flagged directions. When all three directions are flagged, the robot backs up for 0.15 meters then rotates to a different direction.
Our implementation includes straightforward search and chase algorithms. The search state is activated when no objects or targets are detected in its surroundings. During the search state, the controller guides the robot to move forward. Upon receiving a target position, the chase state initiates. In this state, the robot rotates to align the target angle and then moves forward. Finally, the object avoidance state takes precedence over these two states. This state prioritizes steering the robot away from obstacles to ensure its safety.
- Accuracy
- The existing human detection models, such as Yolo, do not work well for our use case because the camera is placed close to the ground.
- Advantages
- Realiable against blurring
- Faster than neural network-based human detections
- Disadvantages
- False-positives on red objects around the field
- Hardware contraints
- Narrow field of view of the camera (73 degrees)
- Depth camera distance limitation (> 0.6 meters)
- Capable of detecting and avoiding walls with high accuracy
- When it comes to relatively smaller objects, such as legs of chairs
We analyzed the state transition with (https://uppaal.org/)
In the UPPAL robot diagram, a 'tagged' state indicates when the robot successfully tags a player within 0.6 meters. Transition between all states is possible except from the initial state. A hierarchical state machine with an upper layer for object avoidance is proposed for future clarity and safety.
Our objective was achieved!
