For my senior year project I'm making a robot that is guided by vision. To keep things simple, the robot itself is a quad rover.
The rover should detect pre-decided objects in its field of view, move up to the object, and grip it.
High-level algorithm
 |
| 1. DETECTION: Detect the object (the window made by black lines represents the physical center of the robot) |
 |
| 2. CENTERING: Rotate the robot to the left or right so that the object comes right in front of the robot |
 |
| 3. DISTANCE: Measure the distance to the object using stereo imaging |
 |
| 4. MOVE FORWARD AND GRIP: Open the gripper, move forward and close the gripper when the object is reached. While moving forward, constantly check if the object is in the central window - if it slips out, stop and correct the course. |
How the system works
Since I do not have a powerful enough embedded processor, I use my laptop computer. The Arduino and the stereo camera must be connected to the laptop computer for the robot to work. The Arduino drives motors while the laptop computer processes frames from the camera and takes decisions. I have programmed them such that the Arduino acts as a client to the laptop computer's server.
- This code runs on the computer. It -
- Detects the object using SURF. For details see my previous post on SURF.
- When the object is in front of the robot (inside the window made by the two black lines) the program also measures distance in centimeters to the object using stereo imaging. For details see my previous post on the topic.
- When the Arduino asks (by sending a '?' through the USB line), the program makes one of these 5 decisions. It communicates this decision to the Arduino by generating a corresponding character:
- Forward - 'w'
- Left - 'a'
- Right - 'd'
- Open gripper - 'o'
- Close gripper - 'c'
- Along with these commands it also sends a character, the value of which determines the interval for which that action is to be done. This is how the intervals are determined -
- For left and right, I conducted trials by moving the motors for 100 ms repeatedly and nothing how much the detected object shifts in pixels. Averaging the results, I got the relation 1 ms = 3 pixels. Now, when the object is detected, I determine the difference in pixels between object center and window center, convert this difference to milliseconds, write that to a character and send it via USB.
- Similarly for forward, I conducted trials that gave me the relation 1 s = 3 cm. Rest of the process remains the same.
- Asks the computer for a command (when it has nothing to do) by sending a '?' character over USB.
- As soon as it receives a command, it sets its output pins (which are connected to the motor driver boards) to particular HIGH or LOW values so that appropriate polarities of the battery are connected to motor terminals and they move to execute the command. For example, if the command received is 'a' (left), the two motors on the right side are to be moved forward and the two on the left side are to be moved backward.
Hardware
- The four motors that drive wheels are 10 rpm 12V 300mA DC motors. The low rpm motors have two advantages
- They give high torque
- Precise control of the robot is possible due to the slow speed
Electronics
- Arduino Uno
- Two Dual H-bridge boards as motor drivers. You can get these from Robokits. This is an excellent tutorial for controlling DC motors through H bridges.
- Capacitors as lowpass filters to remove noise generated by motor back EMF. I used one 47 uF electrolytic capacitor across the battery terminals and Arduino supply terminals each. For the motors I used a 2nF ceramic capacitor across the terminals and a 1 nF ceramic capacitor between each terminal and motor metal body. It is important to use ceramic capacitors only for the motors.
Happy roboting :)