def should_defending(self):
ball = self.info.ball
car = self.info.my_car
our_goal = self.info.my_goal.center
car_to_ball = ball.pos - car.pos
in_front_of_ball = distance_2d(ball.pos, our_goal) < distance_2d(
car.pos, our_goal)
backline_intersect = line_backline_intersect(
self.info.my_goal.center[1], vec2(car.pos), vec2(car_to_ball))
return in_front_of_ball and abs(backline_intersect) < 2000
def should_defend(self):
"""Method which returns a boolean regarding whether we should defend or not"""
ball = self.info.ball
car = self.info.my_car
car_to_ball = ball.position - car.position
in_front_of_ball = self.in_front_off_ball
backline_intersect = line_backline_intersect(self.my_goal.center[1],
vec2(car.position),
vec2(car_to_ball))
return (in_front_of_ball
and abs(backline_intersect) < 2000) or self.conceding
def should_dodge(agent):
""""Method that checks if we should dodge"""
car = agent.info.my_car
their_goal = agent.their_goal
close_to_goal = distance_2d(car.position, their_goal.center) < 4000
aiming_for_goal = abs(line_backline_intersect(
their_goal.center[1], vec2(car.position), vec2(car.forward()))) < 850
bot_to_target = agent.info.ball.position - car.position
local_bot_to_target = dot(bot_to_target, agent.info.my_car.orientation)
angle_front_to_target = math.atan2(local_bot_to_target[1], local_bot_to_target[0])
close_to_ball = norm(vec2(bot_to_target)) < 750
good_angle = abs(angle_front_to_target) < math.radians(15)
return close_to_ball and close_to_goal and aiming_for_goal and good_angle
def should_dodge(agent):
car = agent.info.my_car
their_goal = agent.info.their_goal
close_to_goal = distance_2d(car.pos, their_goal.center) < 4000
aiming_for_goal = abs(
line_backline_intersect(their_goal.center[1], vec2(car.pos),
vec2(car.forward()))) < 850
bot_to_target = agent.info.ball.pos - car.pos
local_bot_to_target = dot(bot_to_target, agent.info.my_car.theta)
angle_front_to_target = math.atan2(local_bot_to_target[1],
local_bot_to_target[0])
close_to_ball = norm(vec2(bot_to_target)) < 850
good_angle = math.radians(-10) < angle_front_to_target < math.radians(10)
return close_to_ball and close_to_goal and aiming_for_goal and good_angle
def should_defending(self):
return False # Don't.
"""Method which returns a boolean regarding whether we should defend or not"""
ball = self.info.ball
car = self.info.my_car
our_goal = self.my_goal.center
car_to_ball = ball.location - car.location
in_front_of_ball = distance_2d(ball.location, our_goal) < distance_2d(
car.location, our_goal)
backline_intersect = line_backline_intersect(self.my_goal.center[1],
vec2(car.location),
vec2(car_to_ball))
return (in_front_of_ball
and abs(backline_intersect) < 2000) or self.conceding
def shooting_target(agent):
""""Method that gives the target for the shooting strategy"""
ball = agent.info.ball
car = agent.info.my_car
car_to_ball = ball.location - car.location
backline_intersect = line_backline_intersect(agent.their_goal.center[1],
vec2(car.location),
vec2(car_to_ball))
if abs(backline_intersect) < 700:
goal_to_ball = normalize(car.location - ball.location)
error = 0
else:
# Right of the ball
if -500 > backline_intersect:
target = agent.their_goal.corners[3] + vec3(400, 0, 0)
# Left of the ball
elif backline_intersect > 500:
target = agent.their_goal.corners[2] - vec3(400, 0, 0)
goal_to_ball = normalize(ball.location - target)
# Subtract the goal to car vector
difference = goal_to_ball - normalize(car.location - target)
error = cap(abs(difference[0]) + abs(difference[1]), 0, 5)
goal_to_ball_2d = vec2(goal_to_ball[0], goal_to_ball[1])
test_vector_2d = dot(rotation(0.5 * math.pi), goal_to_ball_2d)
test_vector = vec3(test_vector_2d[0], test_vector_2d[1], 0)
distance = cap(
(40 + distance_2d(ball.location, car.location) * (error**2)) / 1.8, 0,
4000)
location = ball.location + vec3(
(goal_to_ball[0] * distance), goal_to_ball[1] * distance, 0)
# this adjusts the target based on the ball velocity perpendicular
# to the direction we're trying to hit it
multiplier = cap(distance_2d(car.location, location) / 1500, 0, 2)
distance_modifier = cap(
dot(test_vector, ball.velocity) * multiplier, -1000, 1000)
location += vec3(test_vector[0] * distance_modifier,
test_vector[1] * distance_modifier, 0)
# another target adjustment that applies if the ball is close to the wall
extra = 3850 - abs(location[0])
if extra < 0:
location[0] = cap(location[0], -3850, 3850)
location[1] = location[1] + (-sign(agent.team) * cap(extra, -800, 800))
return location
def shooting_target(agent):
ball = agent.info.ball
car = agent.info.my_car
car_to_ball = ball.pos - car.pos
backline_intersect = line_backline_intersect(
agent.info.their_goal.center[1], vec2(car.pos), vec2(car_to_ball))
if -500 < backline_intersect < 500:
goal_to_ball = normalize(car.pos - ball.pos)
error = cap(distance_2d(ball.pos, car.pos) / 1000, 0, 1)
else:
# Right of the ball
if -500 > backline_intersect:
target = agent.info.their_goal.corners[3] + vec3(400, 0, 0)
# Left of the ball
elif 500 < backline_intersect:
target = agent.info.their_goal.corners[2] - vec3(400, 0, 0)
goal_to_ball = normalize(ball.pos - target)
goal_to_car = normalize(car.pos - target)
difference = goal_to_ball - goal_to_car
error = cap(abs(difference[0]) + abs(difference[1]), 1, 10)
goal_to_ball_2d = vec2(goal_to_ball[0], goal_to_ball[1])
test_vector_2d = dot(rotation(0.5 * math.pi), goal_to_ball_2d)
test_vector = vec3(test_vector_2d[0], test_vector_2d[1], 0)
distance = cap((40 + distance_2d(ball.pos, car.pos) * (error**2)) / 1.8, 0,
4000)
location = ball.pos + vec3(
(goal_to_ball[0] * distance), goal_to_ball[1] * distance, 0)
# this adjusts the target based on the ball velocity perpendicular to the direction we're trying to hit it
multiplier = cap(distance_2d(car.pos, location) / 1500, 0, 2)
distance_modifier = cap(
dot(test_vector, ball.vel) * multiplier, -1000, 1000)
modified_vector = vec3(test_vector[0] * distance_modifier,
test_vector[1] * distance_modifier, 0)
location += modified_vector
# another target adjustment that applies if the ball is close to the wall
extra = 3850 - abs(location[0])
if extra < 0:
location[0] = cap(location[0], -3850, 3850)
location[1] = location[1] + (-sign(agent.team) * cap(extra, -800, 800))
return location
def step(self):
""""Gives output for the dribbling strategy"""
# direction of ball relative to center of car (where should we aim)
# direction of ball relative to yaw of car (where should we aim verse where we are aiming)
local_bot_to_ball = dot(self.ball.position - self.car.position,
self.car.orientation)
angle_front_to_ball = math.atan2(local_bot_to_ball[1],
local_bot_to_ball[0])
# distance between bot and ball
distance = distance_2d(self.car.position, self.ball.position)
# direction of ball velocity relative to yaw of car (which way the ball is moving verse which way we are moving)
if velocity_2d(self.ball.velocity) < 1e-10:
angle_car_forward_to_ball_velocity = 0
else:
angle_car_forward_to_ball_velocity = angle_between(
z_0(self.car.forward()), z_0(self.ball.velocity))
# magnitude of ball_bot_angle (squared)
ball_bot_diff = (self.ball.velocity[0]**2 +
self.ball.velocity[1]**2) - (self.car.velocity[0]**2 +
self.car.velocity[1]**2)
# p is the distance between ball and car
# i is the magnitude of the ball's velocity (squared) the i term would normally
# be the integral of p over time, but the ball's velocity is essentially that number
# d is the relative speed between ball and car
# note that bouncing a ball is distinctly different than balancing something that doesnt bounce
# p_s is the x component of the distance to the ball
# d_s is the one frame change of p_s, that's why p_s has to be global
# we modify distance and ball_bot_diff so that only the component along the car's path is counted
# if the ball is too far to the left, we don't want the bot to think it has to drive forward
# to catch it
distance_y = math.fabs(distance * math.cos(angle_front_to_ball))
distance_x = math.fabs(distance * math.sin(angle_front_to_ball))
# ball moving forward WRT car yaw?
forward = False
if math.fabs(angle_car_forward_to_ball_velocity) < math.radians(90):
forward = True
# first we give the distance values signs
if forward:
d = ball_bot_diff
i = (self.ball.velocity[0]**2 + self.ball.velocity[1]**2)
else:
d = -ball_bot_diff
i = -(self.ball.velocity[0]**2 + self.ball.velocity[1]**2)
if math.fabs(math.degrees(angle_front_to_ball)) < 90:
p = distance_y
else:
p = -1 * distance_y
# this is the PID correction. all of the callibration goes on right here
# there is literature about how to set the variables but it doesn't work quite the same
# because the car is only touching the ball (and interacting with the system) on bounces
# we run the PID formula through tanh to give a value between -1 and 1 for steering input
# if the ball is lower we have no velocity bias
bias_v = 600000 # 600000
# just the basic PID if the ball is too low
if self.ball.position[2] < 120:
correction = np.tanh((20 * p + .0015 * i + .006 * d) / 500)
# if the ball is on top of the car we use our bias (the bias is in velocity units squared)
else:
correction = np.tanh(
(20 * p + .0015 * (i - bias_v) + .006 * d) / 500)
# makes sure we don't get value over .99 so we dont exceed maximum thrust
self.controls.throttle = correction * .99
# anything over .9 is boost
if correction > .99:
self.controls.boost = True
else:
self.controls.boost = False
# this is the PID steering section
# p_s is the x component of the distance to the ball (relative to the cars direction)
# d_s is the on frame change in p_s
# we use absolute value and then set the sign later
d_s = math.fabs(self.p_s) - math.fabs(distance_x)
self.p_s = math.fabs(distance_x)
# give the values the correct sign
if angle_front_to_ball < 0:
self.p_s = -self.p_s
d_s = -d_s
# d_s is actually -d_s ...whoops
d_s = -d_s
max_bias = 35
backline_intersect = line_backline_intersect(self.goal.center[1],
vec2(self.car.position),
vec2(self.car.forward()))
if abs(backline_intersect) < 1000 or self.ball.position[2] > 200:
bias = 0
# Right of the ball
elif -850 > backline_intersect:
bias = max_bias
# Left of the ball
elif 850 < backline_intersect:
bias = -max_bias
# the correction settings can be altered to change performance
correction = np.tanh((100 * (self.p_s + bias) + 1500 * d_s) / 8000)
# apply the correction
self.controls.steer = correction