怎么用代码去写一个场景，用dreamview加载出来

[lsm2842035890]

我想生成一个场景，怎么用代码去创造出来
比如可能程序流程是这样的：①加载地图②放置egovehicle③设置车辆起始点④设置静态障碍物⑤开启一个容器运行场景
问题：有没有代码模板来生成场景，而不是在仿真平台场景编辑中手动设置

[YuqiHuai]

You can check out projects like BehAVExplor, AV-Fuzzer that use code to specify LGSVL scenario. (There are other LGSVL-based implementations but those two are easier to run in my opinion.) I also have some projects that specify SimControl scenarios (scenoRITA, DoppelTest).

I tried to implement scenoRITA in a way that (hopefully) makes it easy to specify SimControl scenario with code. There are still many deficiencies, but you should be able to specify a scenario with a static obstacle and do what you said automatically via code like this

ego = EgoCar(
    initial_position=PositionEstimate(lane_id="lane_27", s=35),
    final_position=PositionEstimate(lane_id="lane_29", s=10),
)
obstacle = Obstacle(
    id=1,
    initial_position=ObstaclePosition(lane_id="lane_29", index=0),
    final_position=ObstaclePosition(lane_id="lane_29", index=0),
    type=ObstacleType.VEHICLE,
    speed=10,
    width=1.5,
    length=2,
    height=1.5,
    motion=ObstacleMotion.STATIC,
)
scenario = Scenario(
    generation_id=0,
    scenario_id=0,
    ego_car=ego,
    obstacles=[obstacle]
)
# ......
ctn = ApolloContainer(APOLLO_ROOT, f"{PROJECT_NAME}_dev_start")
ctn.start_container(verbose=True)
ctn.start_dreamview()
print(f"Dreamview running at {ctn.dreamview_url}")
# ......
run_scenario(ctn, in_docker_path, in_docker_output, scenario_length, initial_x, initial_y, initial_heading)

In DoppelTest, you can specify a scenario with multiple Apollo instances via code

x = Scenario(
    ad_section=ADSection(
        [
            ADAgent(['lane_25', 'lane_19'], 105, 40, 0),
            ADAgent(['lane_25', 'lane_19'], 115, 40, 0),
            ADAgent(['lane_25', 'lane_19'], 125, 40, 0),
        ]
    ),
    pd_section=PDSection([]),
    tc_section=TCSection.get_one())

But since I implemented DoppelTest way earlier, it lacks implementation for automatically installing Apollo and running Apollo the proper way (i.e., as a non-root user).

I am happy to answer any questions if you have trouble making them work for your purpose.

[lsm2842035890]

hi,YuqiHuai,之前我跑过你的DoppelTest项目非常好用，我非常想跟您学习一下如何使用apollo来coding场景的，您之前是自己研究apollo源码吗，比如把这个值传入到某个函数里，开启容器就可以运行设置的这个场景了x = Scenario(
ad_section=ADSection(
[
ADAgent(['lane_25', 'lane_19'], 105, 40, 0),
ADAgent(['lane_25', 'lane_19'], 115, 40, 0),
ADAgent(['lane_25', 'lane_19'], 125, 40, 0),
]
),
pd_section=PDSection([]),
tc_section=TCSection.get_one())
这个是怎么发现的，一步一步debug源码么，有教程讲过这个嘛~

[lilejin322]

是的，我也强烈推荐Yuqi的DoppelTest工作，我们将该项目作为基线进行了稍微的修改，得到了更多有趣的结果，项目的可玩性很高

[YuqiHuai]

谢谢。是自己研究的代码，应该目前没有关于这些的具体教程，当时主要是看着LGSVL的代码做的这个，没用LGSVL因为LGSVL不支持多个EGO car模式下的Ground Truth Perception。Scenario那一部分都是当时自己定义的框架，这部分其实和Apollo自身的代码关系不大，就是看你想怎么定义场景，比如在scenoRITA中又换了另一种代码区定义场景。DoppelTest中和Apollo的交互主要是在这一部分，通过Apollo的CyberRT发送Routing, TrafficLightDetection,和PerceptionObstacles, 是研究了LGSVL的CyberRT Bridge然后做的。

[daohu527]

I think this issue can be discussed in discuss