def simulate(self, bot) -> vec3: # print('simulate intercept') # Init vars c = Car(bot.game.my_car) b = Ball(bot.game.ball) t = vec3(bot.target) intercept = self.location dt = 1.0 / 60.0 hit = False min_error = None # Drive towards intercept (moving in direction of c.forward()) c.rotation = look_at(intercept, c.up()) direction = normalize(intercept - c.location) #c.forward() advance_distance = norm(intercept - c.location) - c.hitbox( ).half_width[0] - b.collision_radius translation = direction * advance_distance sim_start_state: ThrottleFrame = BoostAnalysis().travel_distance( advance_distance, norm(c.velocity)) c.velocity = direction * sim_start_state.speed c.location += translation c.time += sim_start_state.time bot.ball_predictions = [vec3(b.location)] while b.time < c.time: b.step(dt) bot.ball_predictions.append(vec3(b.location)) # print(c.time, b.time) # print(c.location, b.location) # Simulate the collision and resulting for i in range(60 * 3): c.location += c.velocity * dt b.step(dt, c) # Check if we hit the ball yet if norm(b.location - c.location) < (c.hitbox().half_width[0] + b.collision_radius) * 1.05: hit = True # print('hit') # Measure dist from target error = t - b.location if hit and (min_error == None or norm(error) < norm(min_error)): min_error = error # Record trajectory bot.ball_predictions.append(vec3(b.location)) if not hit: return None return min_error
def calculate(car: Car, ball: Ball, target: vec3, ball_predictions=None): # Init vars b = Ball(ball) dt = 1.0 / 60.0 # Generate predictions of ball path if ball_predictions is None: ball_predictions = [] for i in range(60 * 5): b.step(dt) ball_predictions.append(vec3(b.location)) # Gradually converge on ball location by aiming at a location, checking time to that location, # and then aiming at the ball's NEW position. Guaranteed to converge (typically in <10 iterations) # unless the ball is moving away from the car faster than the car's max boost speed intercept = Intercept(b.location) intercept.purpose = 'ball' intercept.boost = True intercept_ball_position = vec3(b.location) collision_achieved = False last_horizontal_error = None last_horizontal_offset = None i = 0 max_tries = 101 analyzer = BoostAnalysis() if intercept.boost else ThrottleAnalysis() while i < max_tries: i += 1 fake_car = Car(car) direction = normalize(intercept.location - car.location) fake_car.rotation = look_at(direction, fake_car.up()) for t in range(60 * 5): # Step car location with throttle/boost analysis data # Not super efficient but POITROAE frame = analyzer.travel_time(dt, norm(fake_car.velocity)) # print('in 1 frame I travel', frame.time, frame.distance, frame.speed) fake_car.location += direction * frame.distance fake_car.velocity = direction * frame.speed fake_car.time += dt ball_location = ball_predictions[t] # Check for collision p = closest_point_on_obb(fake_car.hitbox(), ball_location) if norm(p - ball_location) <= ball.collision_radius: direction_vector = p - (fake_car.location - normalize( fake_car.forward()) * 13.88) # octane center of mass direction_vector[2] = 0 target_direction_vector = target - ball_location target_direction_vector[2] = 0 intercept_ball_position = ball_location direction = atan2(direction_vector[1], direction_vector[0]) ideal_direction = atan2(target_direction_vector[1], target_direction_vector[0]) horizontal_error = direction - ideal_direction # intercept.location = vec3(ball_location) # intercept.time = fake_car.time # return intercept # Now descend the hit direction gradient # Kick off the gradient descent with an arbitrary seed value if last_horizontal_error is None: last_horizontal_error = horizontal_error last_horizontal_offset = 0 if horizontal_error > 0: horizontal_offset = 25 else: horizontal_offset = 25 intercept.location = ball_location - normalize( fake_car.left()) * horizontal_offset break # Recursive case of gradient descent if horizontal_offset == last_horizontal_offset: gradient = 0 else: gradient = (horizontal_error - last_horizontal_error ) / (horizontal_offset - last_horizontal_offset) if gradient == 0: predicted_horizontal_offset = horizontal_offset else: predicted_horizontal_offset = horizontal_offset - horizontal_error / gradient # Base case (convergence) if abs(gradient) < 0.0005: print(f'convergence in {i} iterations') print(f'gradient = {gradient}') print( f'last_horizontal_offset = {last_horizontal_offset}' ) print(f'direction = {degrees(direction)}') print(f'ideal direction = {degrees(ideal_direction)}') print(f'target = {target}') print(f'ball_location = {ball_location}') return intercept # Edge case exit: offset maxed out max_horizontal_offset = car.hitbox( ).half_width[1] + ball.collision_radius if predicted_horizontal_offset > max_horizontal_offset: predicted_horizontal_offset = max_horizontal_offset elif predicted_horizontal_offset < -max_horizontal_offset: predicted_horizontal_offset = -max_horizontal_offset last_horizontal_offset = horizontal_offset last_horizontal_error = horizontal_error horizontal_offset = predicted_horizontal_offset # Return the latest intercept location and continue descending the gradient intercept.location = ball_location - normalize( fake_car.left()) * predicted_horizontal_offset print(f'iteration {i}') print(f'gradient = {gradient}') print(f'horizontal_offset = {horizontal_offset}') print(f'horizontal_error = {degrees(horizontal_error)}') # print(f'ideal direction = {degrees(ideal_direction)}') break # Check for arrival if norm(fake_car.location - intercept.location) < ball.collision_radius / 2: intercept.location = ball_location break if i >= max_tries: print( f'Warning: max tries ({max_tries}) exceeded for calculating intercept' ) return intercept
def get_car_front_center(car: Car): return car.location + normalize(car.forward()) * car.hitbox( ).half_width[0] + normalize(car.up()) * car.hitbox().half_width[2]