| Abstract |
The directed locomotion for the modular robot is one of the challenges of robotics. Although directed locomotion for a certain and simple modular robot can be designed manually, it remains a difficult task. I want to present the works about directed locomotion for the evolutionary robots with arbitrary shape. We divide this topic to two subjects. 1) How robots recognize the direction (other robots)? 2) how the robots move to a direction? In this presentation, I mainly talk about the first subject and mention few works and more plan about the second subject. We adapted the camera as the eye of the robot to recognize the other robots. The challenge of this subject is that recognizing the direction must be working on low performance embedded system. We proposed a solution that combines the colour feature and SVM+HOG. The colour feature was used to find the object-like regions of robots quickly and accurately, and discard the background regions. Then, HOG+SVM is used to recognize the object-like regions. This solution is higher precision and faster more than 67 times than the original HOG+SVM. It was implemented to recognize the other robot on the Raspberry PI 2 embedded a 900 MHz ARM CPU at more than 20fps. For the next step, we will research on how the evolutionary robots move to other robots. |