• Figure 9-5: Two- Point Crossover Method for Template ‘A’ 9.2 Training algorithm and simulation results
  • Ultra fast cnn based Hardware Computing Platform Concepts for adas visual Sensors and Evolutionary Mobile Robots




    Download 3,22 Mb.
    Pdf ko'rish
    bet65/81
    Sana16.05.2024
    Hajmi3,22 Mb.
    #238917
    1   ...   61   62   63   64   65   66   67   68   ...   81
    Bog'liq
    Alireza Fasih

    Figure 
    9-5: Two-
    Point Crossover Method for Template ‘A’
     
    9.2
     
    Training algorithm and simulation results 
    One of the most important parts of this research is simulating both robot and 
    environment. Some authors have implemented the robot and a virtual world by simulation 
    of dynamic rigid bodies [152, 153]. The robot which depends on the physical parameters is 
    implemented in a specific environment. Each part of the robot has a mass, a center of mass, 
    an elasticity parameter, and both dynamic and static friction coefficients. Figure 9-6 shows 
    the implementation of a “snake robot” made up of joints with 2 degrees of freedom. The
    hinges do not have any limitation in rotation. Nevertheless, applying limitation in the 
    rotation range is possible for each joint separately. According to Figure 9-3, each column of 
    the CNN processor is connected to robot actuators. Hence, each actuator of the robot 
    should be connected to one of the columns separately. Since the robot actuators’ response
    time is not equal for all of them we do assume/take the maximum delay for sending the 
    wave on the robot actuators. This delay interval is essential for the robot 
    locomotion/movement. The goal of the learning process is finding optimum templates for 
    moving the robot according to our desires. Finding these templates for a specific 
    movement mechanism/pattern is essential and suitable for the use in a multi layer tasks 
    manager or controlling unit. We are able to use these templates for a low level robotics 
    activity. When a high level controller sends commands to the robot for performing a 


     
    105 
    specific task another controller needs to manage some low level skills like running, 
    turning, jumping and so on, which are necessary to ensure the realization of the high level 
    task [142, 143, 154, 155]. Therefore, by understanding some robot properties the high 
    level task management is very simple in the high level controller. In the above referenced 
    evaluation, authors have tried to find lateral undulation locomotion for a snake robot. Each 
    hinge has two degree of freedom (2-DOF) and can turn in 2 directions. With the method 
    based on genetic algorithms, an optimum template is obtained to make the robot moving 
    or acting according to our desires. The most important point in this learning method is that 
    we don’t predefine any robot kinematics for movement/locomotion in the fitness function.
    The fitness function is a simple and important function which defines the robot behavior in 
    the environment. Complicated rules and equations in the fitness function cannot improve 
    the robot behavioral performance; a simple definition can result to a best robot behavior. 
    Equation 9-2, define the fitness function used for the snake robot lateral undulation 
    locomotion shown in Figure 9-6. 
    (9-2.a) 
    (9-2.b) 
    (9-2.c) 
    (9-2.d) 
    The term 
    ‘AVG’
    denotes the mean distances between parts and the 
    ‘x’
    axis. The term 
    ‘L
    i

    denotes the distance between the 
    i’th
    Download 3,22 Mb.
    1   ...   61   62   63   64   65   66   67   68   ...   81




    Download 3,22 Mb.
    Pdf ko'rish

    Bosh sahifa
    Aloqalar

        Bosh sahifa



    Ultra fast cnn based Hardware Computing Platform Concepts for adas visual Sensors and Evolutionary Mobile Robots

    Download 3,22 Mb.
    Pdf ko'rish