During the simulation BeamNG shows on the console that some control messages miss a byte or have an extra byte. This in turns results in the messages to be discarded and the control commands to get lost.

After joining a running instance of BeamNG, BeamNGpy should tell me if a scenario is already loaded and running

In many occasions, it is important to know the exact time of the simulation. A timer/watch or similar sensor might do it

During the experiments, especially to cut short on execution time, it would be ideal to have a way to force the state of the vehicles (assuming this is possible). For example, changing the position and rotation of the vehicles (Note: #10 enabled force positioning and rotation of vehicles, but this feature disappeared in some release after v0.4), or its speed/acceleration.

During the simulation, there might a need to change some attributes of the vehicles. For example, their color, their components (wheels, spoilers, etc). This can be used to create rich driving simulations while keeping the overall computational cost low (to some extent, the traffic plugin does this). As @pascale-commits says, while spawning a vehicle, color attributes can be defined on-the-fly, which is good; however, I wonder if there's the possibility to change some attributes of cars also without respawning them.

Why this might be important/relevant?

Because the user does not have to "recompute" or respawn the vehicle in the exact condition every time just to change
an attribute that does not impact the physics of the vehicle itself (i.e., color). Said in other words, the user should not do the work of the simulator and replace moving vehicles on the fly.

What attributes can be effectively (easily) changed of a vehicle?

I understand that color is one of such attributes, but what about skins, the color of wheels/rims, windshield, etc.? I assume that any other modification will impact somehow accurately simulating the physics (i.e., the aerodynamics of the car, type/size/material of the wheels, etc.).

have a 3d point (x, y ,z) from Vehicle.get_bbox() .
how can I get the associated point at camera pixel coord.
Or how can I get the camera model and parameters, I can do the compute myself.

BeamNG should provide a method (Programmatic way) to add a static object in the scenario. For instance, Traffic signals.

The ScenarioObject class exists but there is no example of it. It will be great if BeamNG provides the explanation of the parameters to ScenarioObject Class.

For checking that the ego car gets to destination having a sensor which retrieve the relative position of the car w.r.t. a given way point might be useful to have.

The various widgets, e.g., speedometer, that are usually shown on the screen might confuse the sefl-driving AI. Hence, they should probably be remove in the frames dumped during the run.

A workaround is to manually disable them once

Current researchadapter is hardcoded to run on 64256, desired port needs to be passed as a parameter to BeamNG.drive and read client-side.

When collecting data to train AI, one way to improve the data set is to collect images from a slightly different position of the camera and slightly different rotations. This might be achieved by declaring a number of Camera components, or maybe by pausing the simulation and taking all the screenshots with the same camera sensor (but under different configurations), however, a nice meta-sensor/sensor that gets a list of configuration might be easier to use? Such a sensor might be also configured to generate random pose/rotations of the camera while the camera points towards the same point?

Hi,

How to switch on the dip headlights of the vehicle automatically through code, I am not able to find any function in documentation.

At the moment, only one BeamNGpy instance can be used to connect to a simulation. It would be ideal if, at least for vehicles, one can use multiple sockets/end-points, such that one BeamNG instance might be used to control the scenario while the others control the vehicles?

When creating road objects, unless one forces overObjects=1, the roads are not correctly visualized.

#37

The vehicle state contains the physical properties of the car, which can be seen as the ground truth from the simulation. Some sensors read this state to compute their measurements, but not all of them. Therefore, polling those sensors does not update (or create) the state of the vehicle which is inaccurate.

The BeamNG throws an error when I provide the set of triples in the list for the vehicle.

`try:

    bng.load_scenario(scenario)
    bng.start_scenario()
    line = [(0,0,0.163609),(25,300, 0.163609)]
    vehicle.ai_set_line(line=line)
    vehicle.ai_set_speed(6.944, mode='set')
    vehicle.ai_drive_in_lane(True)
    input('Press enter when done...')
finally:
    bng.close()`

The sample output in the console is attached.

It will be great if you can correct me if I am doing something wrong.

A programmatic way to configure how to grab the frames from BeamNG (camera position, FOV, etc) would be ideal.

Would it be possible to query a running simulation from BeamNGpy? For example, getting how many (and which) cars are running, etc.

When I try to connect to beamNG research, it connects for first time and then disconnect

here is the log

029.612|I|libbeamng.lua.V.ResearchVE|Trying to connect to: localhost:64257
031.627|I|libbeamng.lua.V.ResearchVE|Connected!
031.630|I|GELua.util_researchGE.ResearchGE|Setting engine flags!
031.662|D|GELua.scenario_scenarios.scenarios|changeState: running
031.663|D|GELua.core_camera.|Camera switched to "orbit"
031.712|E|GELua.core_levels.levels|info.json missing: /levels/garage/info.json
031.713|E|GELua.core_levels.levels|No entry point for level found: /levels/garage. Ignoring level.
031.714|E|GELua.core_levels.levels|info.json missing: /levels/GridMap/info.json
031.715|E|GELua.core_levels.levels|No entry point for level found: /levels/GridMap. Ignoring level.
031.819|E|GELua.ResearchCom|Error reading from socket: closed
031.874|E|libbeamng.lua.V.ResearchCom|Error reading from socket: closed

Currently spaces in vehicle names make the prefab not compile

In the annotation mode of the camera lanes, buildings, participants etc. can be vizualized with predefined colors whereas ProceduralMeshes like Cube, Cone and Cylinder can not.
This would be interesting the same way like annotating buildings.

If you add several Decalroads to the prefab file by using the scenario.add_road function, then they all spawn at (0,0) even though their start and endpoint are different. The same prefab file works fine for BeamNG.research version 1.15. There are no error messages in the BeamNG console.
Prefab file: https://pastebin.com/LXL7Vp7r

I am trying to create multiple layers of the road in BeamNGpy in the scenario. The asphalt road and lane marking lie on the same level (height). Hence, the lane marking is not visible in BeamNG scenario. The API should provide a method to control the road position vertically in GridMap.

Mostly for debugging it woule be nice if BeamNGpy could start BeamNG using the '-console' option on demand

Images collected using the camera sensor while the car is braking shows a red shadow, despite the camera is positioned in front on the car facing the road.

Tools like traffic tool can dynamically spawn cars in running games/simulations. Maybe we can add the same feature to BeamNGpy?

Currently, the API allows to specify driving commands only. This is fine for letting external AI to drive the cars, but it might be not optimal for test generators which need to place the cars in specific position/rotations.

Hi guys,

I am not able to get vehicle bounding box by given bng.get_vehicle_bbox(vehicle2) after starting the scenario.

bng.load_scenario(scenario)
bng.start_scenario()
vehicle.ai_set_line(path1)
vehicle2.ai_set_line(path2)
vehicle3.ai_set_line(path3)
bng.get_vehicle_bbox(vehicle2)
# Following error occurs:

bbox = bng.get_vehicle_bbox(vehicle2)
  File "D:\Beamng_qazi_data\Thesis_final_code\venv\lib\site-packages\beamngpy\beamng.py", line 1130, in get_vehicle_bbox
    resp = self.recv()
  File "D:\Beamng_qazi_data\Thesis_final_code\venv\lib\site-packages\beamngpy\beamng.py", line 241, in recv
    return recv_msg(self.skt)
  File "D:\Beamng_qazi_data\Thesis_final_code\venv\lib\site-packages\beamngpy\beamngcommon.py", line 155, in recv_msg
    length = int(str(length, 'ascii'))

According to the function it should return:
near_bottom_left: Bottom left point of the near rectangle as an (x, y ,z) triplet
• near_bottom_right: Bottom right point of the near rectangle as an (x, y, z) triplet
• near_top_left: Top left point of the near rectangle as an (x, y, z) triplet
• near_top_right: Top right point of the near rectangle as an (x, y, z) triplet
• far_bottom_left: Bottom left point of the far rectangle as an (x, y, z) triplet
• far_bottom_right: Bottom right point of the far rectangle as an (x, y, z) triplet
• far_top_left: Top left point of the far rectangle as an (x, y, z) triplet
• far_top_right: Top right point of the far rectangle as an (x, y, z) triplet

Examples:

• Which cars are in the simulation?
• Which waypoints are in the simulation?

Possibly navigational data of some sort:

• Which streets/roads/etc are in the simulation?

Hello!

Is it possible to create AI traffic from BeamNGpy which could be launched together with a player-controlled car?
Can you share a code example if it is possible?

The API defaults to controlling vehicle 0, this needs to be changeable by the user.

If brake=1.0 the car goes into Reverse mode and drives backwards.

How one could import new elements/models into a scenario?
Should those elements be already available during "scenario generation time" or would it possible to import them on the fly from "file" or "URL"?

Steps to reproduce:
- Start BeamNG with -console, load scenario
- visualize AIControl HUD, change AI mode to "Manual"
- set In Lane to "on"
- set target waypoint (NOTE it is important to set In Lane before the way point otherwise the car will NOT stay in the lane since the beginning)
- change AI mode to "Disabled"

Check that there's no error messages in BeamNG Console.

- Start BeamNG.py and  call:
    - vehicle.ai_set_mode("manual")
    - vehicle.ai_drive_in_lane(True) *I also tried with "on" and "'on'"
    - vehicle.ai_set_waypoint(<waypoint>)

Check that AI Control HUD does not update to reflect the manual mode

     - set In Lane to "on" or "off" 

Check in the BeamNG console that the following error message is printed:

056.717|E|libbeamng.lua.V|=============== Stack Traceback >> START >>
056.718|E|libbeamng.lua.V|(1) main chunk of [string "string"] at line 1
056.718|E|libbeamng.lua.V|--------------- << END <<
056.718|E|libbeamng.lua.V|*** while executing this line: ai.driveInLane("on")
056.719|E|libbeamng.lua.V|*** in chunk: string```
    
 

I just downloaded (via SVN) the BeamNG.research repository.
https://projects.beamng.com/projects/research/files is empty. In a conda enviroment I installed BeamNGpy via pip install beamngpy.

When I try the example code with west_coast_usa, BeamNG starts but after a short time it threws the following error:

055.319|D|GELua.scenario_scenarios.scenarios|should run: false; state: running 055.322|E|GELua.core_levels.levels|info.json missing: /levels/garage/info.json 055.325|E|GELua.core_levels.levels|No entry point for level found: /levels/garage. Ignoring level. 055.442|E|GELua.ResearchCom|Error reading from socket: closed 055.472|E|GELua.core_levels.levels|info.json missing: /levels/garage/info.json 055.473|E|GELua.core_levels.levels|No entry point for level found: /levels/garage. Ignoring level. 055.613|E|libbeamng.lua.V.ResearchCom|Error reading from socket: closed 057.143|I|libbeamng.lua.V.ResearchVE|Trying to connect to: localhost:64257 059.355|I|libbeamng.lua.V.ResearchVE|Could not connect... 060.001|I|GELua.util_researchGE.ResearchGE|Trying to connect to: localhost:64256 062.317|I|GELua.util_researchGE.ResearchGE|Could not connect... 063.386|I|libbeamng.lua.V.ResearchVE|Trying to connect to: localhost:64257 065.533|I|libbeamng.lua.V.ResearchVE|Could not connect... 065.735|I|GELua.util_researchGE.ResearchGE|Trying to connect to: localhost:64256 067.760|I|GELua.util_researchGE.ResearchGE|Could not connect... 074.911|I|GELua.util_researchGE.ResearchGE|Trying to connect to: localhost:64256

What could be the problem?

If I am not mistaken, the semantic annotations provided by BeamNG uses the same color/label for each type of object. This makes it difficult to distinguish cars when they are close to each other (or one occludes the other) and tracking the movement of objects. Maybe we can have each object a different label and some additional metadata attached to the image. For example, color X, element Y, type=car, model=Z, etc. ?

Hi there!
I found that BeamNG could fit well for my research. Now I'm looking for a way to collect data from the game.
In the youtube demonstration from Engineering psychology I found a great example of what i need (1:30-1:35):

I want to export at least:

position
speed
wheel angle
collision between cars (if the car is damaged)

Is there a simple option to export such data to csv?

In some situations calling the following code:

self.vehicle.ai_set_script(self.script)```
results in teleporting the car, which is somehow an unexpected behavior. This might be something related to the "current speed" of the car which is zero at the beginning or it might be something else.

It would be nice to have the heading angle of the vehicle with respect to the road it is traveling on

After joining a running BeamNG instance, BeamNGpy should be able to query which vehicles are there, and taking control over them.

When a Lidar sensor is attached and visualized, the point cloud is visible in the Camera sensor's rendered frames. They shouldn't.

BeamNGpy is the latest from PyPI (1.14.1) and BeamNG.research is the latest one from subversion. However, this is the warning message I get:

BeamNGpy and BeamNG.* version mismatch: BeamNGpy v1.12, BeamNG.* v1.9
Make sure both this library and BeamNG.* are up to date.
Operation will proceed, but some features might not work.
<beamngpy.beamng.BeamNGpy object at 0x000002B1B6F94048>

Is there a newer version of BeamNG.research somewhere else?

I am trying to load and start existing scenarios so I do:

• Start beamng:
  beamng = BeamNGpy('localhost', 64256)
  bng = beamng.open(launch=True)

• Define and (find) the existing scenario:
  scenario1.bng = bng
  scenario1.find(bng)

• Load the scenario:
  bng.load_scenario(scenario1)

• Start the scenario:
  bng.start_scenario()

However, after calling load scenario (and I can see that the scenario is loaded), start scenario blocks.