def steer(self, desired):
        # steer = vector((0, 1))

        if norm(desired) > self.maxspeed:
            desired = normalize(desired) * self.maxspeed
        self.desired = desired
        steer = desired - self.velocity
        steer = normalize(steer) * self.maxforce
        # steer.normalize()

        self.apply(steer)
 def separate(self, vehicles):
     radius = self.size * 2
     sum_vel = vector((0.0, 0.0))
     n = 0
     for other in vehicles:
         d = dist(self.position, other.position)
         if 0 < d < radius:
             n += 1
             diff = normalize(self.position - other.position) / d
             sum_vel += diff
     if n > 0:
         sum_vel /= n
         sum_vel = normalize(sum_vel) * self.maxspeed
         self.steer(sum_vel)
    def wander(self, d=50, r=25, change=0.5):
        center = normalize(self.velocity) * d + self.position
        self.theta += random() * 2 * change - change
        randrad = vector((sin(self.theta), cos(self.theta))) * r
        desired = center + randrad

        self.steer(desired - self.position)
Exemple #4
0
 def __init__(self, path, resolution=None):
     resolution = resolution or path.raduis / 1.5
     super().__init__(resolution=resolution)
     for i in range(self.cols):
         for j in range(self.rows):
             center = vector((int((i + 0.5) * self.resolution),
                              int((j + 0.5) * self.resolution)))
             self.field[i][j] = normalize(path.get_normal(center) -
                                          center)  # * 0.2
 def align(self, vehicles):
     sum_vel = vector((0.0, 0.0))
     n = 0
     for other in vehicles:
         if other is not self:
             sum_vel += other.velocity
             n += 1
     if n > 0:
         sum_vel /= n
         sum_vel = normalize(sum_vel) * self.maxspeed
         # print(self.velocity, sum_vel)
         self.steer(sum_vel)
    def track(self, path, scr=None, debug=False):
        prediction_rate = 25
        future_loc = self.position + normalize(self.velocity) * prediction_rate
        # future_loc = vector((0, 1))
        normal_loc = path.get_normal(future_loc)

        if debug:
            pygame.draw.circle(scr, (255, 128, 255),
                               (int(normal_loc[0]), int(normal_loc[1])), 10)
            pygame.draw.circle(scr, (255, 230, 255),
                               (int(future_loc[0]), int(future_loc[1])), 10)

        if dist(future_loc, normal_loc) > path.raduis:
            self.seek(normal_loc)
 def follow(self, field):
     # desired = field.lookup(self.position + self.velocity)
     desired = normalize(field.lookup(self.position)) * self.maxspeed
     self.steer(desired)