Learning an efficient way to retrieve simulation data is essential in CARLA. This holistic tutorial is advised for both, newcomers and more experienced users. It starts from the very beginning, and gradually dives into the many options available in CARLA.
First, the simulation is initialized with custom settings and traffic. An ego vehicle is set to roam around the city, optionally with some basic sensors. The simulation is recorded, so that later it can be queried to find the highlights. After that, the original simulation is played back, and exploited to the limit. New sensors can be added to retrieve consistent data. The weather conditions can be changed. The recorder can even be used to test specific scenarios with different outputs.
There are some common mistakes in the process of retrieving simulation data. Flooding the simulator with sensors, storing useless data, or struggling to find a specific event are some examples. However, some outlines to this process can be provided. The goal is to ensure that data can be retrieved and replicated, and the simulation can be examined and altered at will.
This tutorial uses the [__CARLA 0.9.8 deb package__](start_quickstart.md). There may be minor changes depending on your CARLA version and installation, specially regarding paths.
The tutorial presents a wide set of options for the differents steps. All along, different scripts will be mentioned. Not all of them will be used, it depends on the specific use cases. Most of them are already provided in CARLA for generic purposes.
However, there are two scripts mentioned along the tutorial that cannot be found in CARLA. They contain the fragments of code cited. This serves a twofold purpose. First of all, to encourage users to build their own scripts. It is important to have full understanding of what the code is doing. In addition to this, the tutorial is only an outline that may, and should, vary a lot depending on user preferences. These two scripts are just an example.
* __tutorial_ego.py__ spawns an ego vehicle with some basic sensors, and enables autopilot. The spectator is placed at the spawning position. The recorder starts at the very beginning, and stops when the script is finished.
* __tutorial_replay.py__ reenacts the simulation that __tutorial_ego.py__ recorded. There are different fragments of code to query the recording, spawn some advanced sensors, change weather conditions, and reenact fragments of the recording.
The full code can be found in the last section of the tutorial. Remember these are not strict, but meant to be customized. Retrieving data in CARLA is as powerful as users want it to be.
Choose a map for the simulation to run. Take a look at the [map documentation](core_map.md#carla-maps) to learn more about their specific attributes. For the sake of this tutorial, __Town07__ is chosen.
Each town is loaded with a specific weather that fits it, however this can be set at will. There are two scripts that offer different approaches to the matter. The first one sets a dynamic weather that changes conditions over time. The other sets custom weather conditions. It is also possible to code weather conditions. This will be covered later when [changing weather conditions](#change-the-weather).
* __To set a dynamic weather__. Open a new terminal and run __dynamic_weather.py__. This script allows to set the ratio at which the weather changes, being `1.0` the default setting.
* __To set custom conditions__. Use the script __environment.py__. There are quite a lot of possible settings. Take a look at the optional arguments, and the documentation for [carla.WeatherParameters](python_api.md#carla.WeatherParameters).
Simulating traffic is one of the best ways to bring the map to life. It is also necessary to retrieve data for urban environments. There are different options to do so in CARLA.
The CARLA traffic is managed by the [Traffic Manager](adv_traffic_manager.md) module. As for pedestrians, each of them has their own [carla.WalkerAIController](python_api.md#carla.WalkerAIController).
CARLA can run a co-simulation with SUMO. This allows for creating traffic in SUMO that will be propagated to CARLA. This co-simulation is bidirectional. Spawning vehicles in CARLA will do so in SUMO. Specific docs on this feature can be found [here](adv_sumo.md).
* With the CARLA server on, run the [SUMO-CARLA synchrony script](https://github.com/carla-simulator/carla/blob/master/Co-Simulation/Sumo/run_synchronization.py).
The traffic generated by this script is an example created by the CARLA team. By default it spawns the same vehicles following the same routes. These can be changed by the user in SUMO.
From now up to the moment the recorder is stopped, there will be some fragments of code belonging to __tutorial_ego.py__. This script spawns the ego vehicle, optionally some sensors, and records the simulation until the user finishes the script.
Vehicles controlled by the user are commonly differenciated in CARLA by setting the attribute `role_name` to `ego`. Other attributes can be set, some with recommended values.
Hereunder, a Tesla model is retrieved from the [blueprint library](bp_library.md), and spawned with a random recommended colour. One of the recommended spawn points by the map is chosen to place the ego vehicle.
__2.__ Set specific attributes for the sensor. This is crucial. Attributes will shape the data retrieved.
__3.__ Attach the sensor to the ego vehicle. __The transform is relative to its parent__. The [carla.AttachmentType](python_api.md#carlaattachmenttype) will determine how the position of the sensor is updated.
__4.__ Add a `listen()` method. This is the key element. A [__lambda__](https://www.w3schools.com/python/python_lambda.asp) method that will be called each time the sensor listens for data. The argument is the sensor data retrieved.
Having this basic guideline in mind, let's set some basic sensors for the ego vehicle.
The [RGB camera](ref_sensors.md#rgb-camera) generates realistic pictures of the scene. It is the sensor with more settable attributes of them all, but it is also a fundamental one. It should be understood as a real camera, with attributtes such as `focal_distance`, `shutter_speed` or `gamma` to determine how it would work internally. There is also a specific set of attributtes to define the lens distorsion, and lots of advanced attributes. For example, the `lens_circle_multiplier` can be used to achieve an effect similar to an eyefish lens. Learn more about them in the [documentation](ref_sensors.md#rgb-camera).
For the sake of simplicity, the script only sets the most commonly used attributes of this sensor.
After setting the attributes, it is time to spawn the sensor. The script places the camera in the hood of the car, and pointing forward. It will capture the front view of the car.
The data is retrieved as a [carla.Image](python_api.md#carla.Image) on every step. The listen method saves these to disk. The path can be altered at will. The name of each image is coded to be based on the simulation frame where the shot was taken.
These sensors retrieve data when the object they are attached to registers a specific event. There are three type of detector sensors, each one describing one type of event.
The data they retrieve will be helpful later when deciding which part of the simulation is going to be reenacted. In fact, the collisions can be explicitely queried using the recorder. This is prepared to be printed.
* __`sensor_tick`__ sets the sensor to retrieve data only after `x` seconds pass. It is a common attribute for sensors that retrieve data on every step.
* __`distance` and `hit-radius`__ shape the debug line used to detect obstacles ahead.
* __`only_dynamics`__ determines if static objects should be taken into account or not. By default, any object is considered.
The script sets the obstacle detector to only consider dynamic objects. If the vehicle collides with any static object, it will be detected by the collision sensor.
The attributes available for these sensors mostly set the mean or standard deviation parameter in the noise model of the measure. This is useful to get more realistic measures. However, in __tutorial_ego.py__ only one attribute is set.
* __`sensor_tick`__. As this measures are not supposed to vary significantly between steps, it is okay to retrieve the data every so often. In this case, it is set to be printed every three seconds.
The script __tutorial_replay.py__, among other things, contains definitions for more sensors. They work in the same way as the basic ones, but their comprehension may be a bit harder.
The [depth camera](ref_sensors.md#depth-camera) generates pictures of the scene that map every pixel in a grayscale depth map. However, the output is not straightforward. The depth buffer of the camera is mapped using a RGB color space. This has to be translated to grayscale to be comprehensible.
In order to do this, simply save the image as with the RGB camera, but apply a [carla.ColorConverter](python_api.md#carla.ColorConverter) to it. There are two conversions available for depth cameras.
* __carla.ColorConverter.Depth__ translates the original depth with milimetric precision.
* __carla.ColorConverter.LogarithmicDepth__ also has milimetric granularity, but provides better results in close distances and a little worse for further elements.
The attributes for the depth camera only set elements previously stated in the RGB camera: `fov`, `image_size_x`, `image_size_y` and `sensor_tick`. The script sets this sensor to match the previous RGB camera used.
The [semantic segmentation camera](ref_sensors.md#semantic-segmentation-camera) renders elements in scene with a different color depending on how these have been tagged. The tags are created by the simulator depending on the path of the asset used for spawning. For example, meshes tagged as `Pedestrians` are spawned with content stored in `Unreal/CarlaUE4/Content/Static/Pedestrians`.
The output is an image, as any camera, but each pixel contains the tag encoded in the red channel. This original image must be converted using __ColorConverter.CityScapesPalette__. New tags can be created, read more in the [documentation](ref_sensors.md#semantic-segmentation-camera).
The [LIDAR sensor](ref_sensors.md#lidar-raycast-sensor) simulates a rotating LIDAR. It creates a cloud of points that maps the scene in 3D. The LIDAR contains a set of lasers that rotate at a certain frequency. The lasers raycast the distance to impact, and store every shot as one single point.
The way the array of lasers is disposed can be set using different sensor attributes.
Other attributes set the way this points are calculated. They determine the amount of points that each laser calculates every step: `points_per_second / (FPS * channels)`.
The point cloud output is described as a [carla.LidarMeasurement]. It can be iterated as a list of [carla.Location] or saved to a _.ply_ standart file format.
The [radar sensor](ref_sensors.md#radar-sensor) is similar to de LIDAR. It creates a conic view, and shoots lasers inside to raycast their impacts. The output is a [carla.RadarMeasurement](python_api.md#carlaradarmeasurement). It contains a list of the [carla.RadarDetection](python_api.md#carlaradardetection) retrieved by the lasers. These are not points in space, but detections with data regarding the sensor: `azimuth`, `altitude`, `sensor` and `velocity`.
The script places the sensor on the hood of the car, and rotated a bit upwards. That way, the output will map the front view of the car. The `horizontal_fov` is incremented, and the `vertical_fov` diminished. The area of interest is specially the height where vehicles and walkers usually move on. The `range` is also changed from 100m to 10m, in order to retrieve data only right ahead of the vehicle.
The callback is a bit more complex this time, showing more of its capabilities. It will draw the points captured by the radar on the fly. The points will be colored depending on their velocity regarding the ego vehicle.
<divstyle="text-align: right"><i>Radar output. The vehicle is stopped at a traffic light, so the static elements in front of it appear in white.</i></div>
The [no-rendering mode](adv_rendering_options.md) can be useful to run an initial simulation that will be later played again to retrieve data. Especially if this simulation has some extreme conditions, such as dense traffic.
### Simulate at a fast pace
Disabling the rendering will save up a lot of work to the simulation. As the GPU is not used, the server can work at full speed. This could be useful to simulate complex conditions at a fast pace. The best way to do so would be by setting a fixed time-step. Running an asynchronous server with a fixed time-step and no rendering, the only limitation for the simulation would be the inner logic of the server.
The same `config.py` used to [set the map](#map-setting) can disable rendering, and set a fixed time-step.
Read the [documentation](adv_synchrony_timestep.md) before messing around with with synchrony and time-step.
### Manual control without rendering
The script `PythonAPI/examples/no_rendering_mode.py` provides an overview of the simulation. It creates a minimalistic aerial view with Pygame, that will follow the ego vehicle. This could be used along with __manual_control.py__ to generate a route with barely no cost, record it, and then play it back and exploit it to gather data.
<divstyle="text-align: right"><i>no_rendering_mode.py working in Town07</i></div>
!!! Note
In this mode, GPU-based sensors will retrieve empty data. Cameras are useless, but other sensors such as detectors will work properly.
---
## Record and retrieve data
### Start recording
The [__recorder__](adv_recorder.md) can be started at anytime. The script does it at the very beginning, in order to capture everything, including the spawning of the first actors. If no path is detailed, the log will be saved into `CarlaUE4/Saved`.
There are many different ways to do this. Mostly it goes down as either let it roam around or control it manually. The data for the sensors spawned will be retrieved on the fly. Make sure to check it while recording, to make sure everything is set properly.
* __Enable the autopilot.__ This will register the vehicle to the [Traffic Manager](adv_traffic_manager.md). It will roam around the city endlessly. The script does this, and creates a loop to prevent the script from finishing. The recording will go on until the user finishes the script. Alternatively, a timer could be set to finish the script after a certain time.
```py
# --------------
# Capture data
# --------------
ego_vehicle.set_autopilot(True)
print('\nEgo autopilot enabled')
while True:
world_snapshot = world.wait_for_tick()
```
* __Manual control.__ Run the script `PythonAPI/examples/manual_control.py` in a client, and the recorder in another one. Drive the ego vehicle around to create the desired route, and stop the recorder when finished. The __tutorial_ego.py__ script can be used to manage the recorder, but make sure to comment other fragments of code.
To avoid rendering and save up computational cost, enable [__no rendering mode__](adv_rendering_options.md#no-rendering-mode). The script `/PythonAPI/examples/no_rendering_mode.py` does this while creating a simple aerial view.
### Stop recording
The stop call is even simpler than the start call was. When the recorder is done, the recording will be saved in the path stated previously.
So far, a simulation has been recorded. Now, it is time to examine the recording, find the most remarkable moments, and work with them. These steps are gathered in the script, __tutorial_replay.py__. The outline is structured in different segments of code commented.
The different queries are detailed in the [__recorder documentation__](adv_recorder.md). In summary, they retrieve data for specific events or frames. Use the queries to study the recording. Find the spotlight moments, and trace what can be of interest.
After the queries, it may be a good idea play some moments of the simulation back, before messing around. It is very simple to do so, and it could be really helpful. Know more about the simulation. It is the best way to save time later.
* __Use the information from the queries.__ Find out the moment and the actors involved in an event, and play that again. Start the recorder a few seconds before the event.
* __Follow different actors.__ Different perspectives will show new events that are not included in the queries.
* __Rom around with a free spectator view.__ Set the `actor_id` to `0`, and get a general view of the simulation. Be wherever and whenever wanted thanks to the recording.
When the recording stops, the simulation doesn't. Walkers will stand still, and vehicles will continue roaming around. This may happen either if the log ends, or the playback gets to the ending point stated.
Gather a list of the important moments, actors and events. Add sensors whenever needed and play the simulation back. The process is exactly the same as before. The script __tutorial_replay.py__ provides different examples that have been thoroughly explained in the [__Set advanced sensors__](#set-advanced-sensors) section. Others have been explained in the section [__Set basic sensors__](#set-basic-sensors).
The recording will recreate the original weather conditions. However, these can be altered at will. This may be interesting to compare how does it affect sensors, while mantaining the rest of events the same.
Get the current weather and modify it freely. Remember that [carla.WeatherParameters](python_api.md#carla.WeatherParameters) has some presets available. The script will change the environment to a foggy sunset.
This can be profitable for the user. For instance, collisions can be forced or avoided by playing back the simulation a few seconds before, and spawning or destroying an actor. Ending the recording at a specific moment can also be useful. Doing so, vehicles may take different paths.
Change the conditions and mess with the simulation. There is nothing to lose, as the recorder grants that the initial simulation can always be reenacted. This is the key to exploit the full potential of CARLA.
Hereunder are the two scripts gathering the fragments of code for this tutorial. Most of the code is commented, as it is meant to be modified to fit specific purposes.
That is a wrap on how to properly retrieve data from the simulation. Make sure to play around, change the conditions of the simulator, experiment with sensor settings. The possibilities are endless.
Visit the forum to post any doubts or suggestions that have come to mind during this reading.
