SUMO与CARLA联合仿真实现遇到的问题求助

核心思路 #

联合仿真的关键是双向状态同步：SUMO负责全局车流的路径规划与状态更新，CARLA负责自车的感知、决策与控制，两者通过固定步长的仿真循环，实时交换车辆位置、速度、航向等核心数据，确保场景一致性。

步骤拆解与实现代码 #

1. 环境与基础准备 #

• 确保版本兼容：推荐使用SUMO 1.16+、CARLA 0.9.14+，避免API不匹配问题
• 安装依赖：SUMO的traci接口（随SUMO安装自带）、CARLA Python客户端包
• 场景地图对齐：用SUMO的netconvert工具将CARLA的OpenDRIVE地图转换为SUMO路网，或反向导出，确保两者坐标系匹配

2. 坐标系转换（关键） #

SUMO采用笛卡尔坐标系，CARLA基于UE4坐标系，需编写转换函数对齐：

import math

def sumo_to_carla(sumo_x, sumo_y, sumo_angle):
    # 根据你的场景调整缩放/偏移系数，示例为反向y轴
    carla_x = sumo_x
    carla_y = -sumo_y
    carla_yaw = sumo_angle  # 若角度方向不一致，需加180或调整正负
    return carla_x, carla_y, carla_yaw

def carla_to_sumo(carla_x, carla_y, carla_yaw):
    sumo_x = carla_x
    sumo_y = -carla_y
    sumo_angle = carla_yaw
    return sumo_x, sumo_y, sumo_angle

3. 仿真启动与同步框架 #

启动SUMO并连接

import traci

# 启动SUMO，设置固定步长0.1s
sumo_cfg_path = "./your_scenario.sumocfg"
sumo_cmd = ["sumo", "-c", sumo_cfg_path, "--step-length", "0.1"]
traci.start(sumo_cmd)

# 将SUMO中的自车设置为受控状态（避免SUMO自动控制）
traci.vehicle.setSpeedMode("ego_vehicle", 0)  # 禁用SUMO的速度控制
traci.vehicle.setLaneChangeMode("ego_vehicle", 0)  # 禁用SUMO的变道控制

启动CARLA并加载地图

import carla

client = carla.Client("localhost", 2000)
client.set_timeout(10.0)
world = client.load_world("Town03")  # 替换为你的场景地图
settings = world.get_settings()
settings.fixed_delta_seconds = 0.1  # 与SUMO步长保持一致
world.apply_settings(settings)

# 生成CARLA自车（需与SUMO中自车ID对应）
blueprint_library = world.get_blueprint_library()
ego_bp = blueprint_library.find("vehicle.tesla.model3")
ego_spawn_point = carla.Transform(carla.Location(x=0, y=0, z=0.5))  # 初始位置与SUMO对齐
ego_vehicle = world.spawn_actor(ego_bp, ego_spawn_point)

4. 双向状态同步 #

SUMO车流同步到CARLA

遍历SUMO中除自车外的所有车辆，更新CARLA中对应车辆的状态：

# 预存CARLA车辆字典，避免重复生成
carla_vehicles = {}

def sync_sumo_to_carla():
    sumo_veh_ids = traci.vehicle.getIDList()
    for veh_id in sumo_veh_ids:
        if veh_id == "ego_vehicle":
            continue
        # 获取SUMO车辆状态
        sumo_pos = traci.vehicle.getPosition(veh_id)
        sumo_speed = traci.vehicle.getSpeed(veh_id)
        sumo_angle = traci.vehicle.getAngle(veh_id)
        # 转换为CARLA坐标
        carla_x, carla_y, carla_yaw = sumo_to_carla(sumo_pos[0], sumo_pos[1], sumo_angle)
        # 生成或更新CARLA车辆
        if veh_id not in carla_vehicles:
            veh_bp = blueprint_library.find("vehicle.tesla.model3")  # 替换为对应车型
            spawn_transform = carla.Transform(
                carla.Location(x=carla_x, y=carla_y, z=0.5),
                carla.Rotation(yaw=carla_yaw)
            )
            carla_veh = world.spawn_actor(veh_bp, spawn_transform)
            carla_vehicles[veh_id] = carla_veh
        else:
            carla_veh = carla_vehicles[veh_id]
            # 更新位置与速度
            target_transform = carla.Transform(
                carla.Location(x=carla_x, y=carla_y, z=0.5),
                carla.Rotation(yaw=carla_yaw)
            )
            carla_veh.set_transform(target_transform)
            carla_veh.set_target_velocity(carla.Vector3D(x=sumo_speed, y=0, z=0))

CARLA自车同步到SUMO

将CARLA中自车的实时状态反馈给SUMO，确保车流能避让自车：

def sync_carla_to_sumo():
    ego_transform = ego_vehicle.get_transform()
    ego_vel = ego_vehicle.get_velocity()
    # 转换为SUMO坐标与状态
    sumo_x, sumo_y, sumo_angle = carla_to_sumo(
        ego_transform.location.x,
        ego_transform.location.y,
        ego_transform.rotation.yaw
    )
    sumo_speed = math.sqrt(ego_vel.x**2 + ego_vel.y**2)
    # 更新SUMO自车状态
    traci.vehicle.setPosition("ego_vehicle", (sumo_x, sumo_y))
    traci.vehicle.setSpeed("ego_vehicle", sumo_speed)
    traci.vehicle.setAngle("ego_vehicle", sumo_angle)

5. 主仿真循环 #

import time

# 替换为你的自车掉头控制逻辑
def get_ego_driving_control():
    # 示例：简单的掉头控制，可替换为PID/强化学习模型
    control = carla.VehicleControl()
    control.throttle = 0.3
    control.steer = 0.5  # 左转掉头，根据场景调整
    return control

try:
    while traci.simulation.getMinExpectedNumber() > 0:
        # 1. 同步SUMO车流到CARLA
        sync_sumo_to_carla()
        # 2. 执行CARLA自车控制
        ego_control = get_ego_driving_control()
        ego_vehicle.apply_control(ego_control)
        # 3. 同步CARLA自车到SUMO
        sync_carla_to_sumo()
        # 4. 推进仿真步
        traci.simulationStep()
        time.sleep(0.1)  # 与步长保持一致
finally:
    # 清理资源
    traci.close()
    ego_vehicle.destroy()
    for veh in carla_vehicles.values():
        veh.destroy()

关键注意事项 #

• 步长严格同步：SUMO与CARLA的仿真步长必须完全一致，否则会出现状态错位
• SUMO自车配置：在SUMO的rou.xml中需将自车标记为controlled="true"，避免SUMO自动干预
• 碰撞规避：若出现车流与自车碰撞，可调整SUMO的carFollowModel参数（如增加安全距离），或优化状态同步的延迟
• 性能优化：避免在每一步都重复生成车辆，用字典缓存已生成的CARLA车辆实例

内容的提问来源于stack exchange，提问作者Kristof Papp