Pedestrian Interaction with (Autonomous) Vehicles

[M-Colley]

According to the documentation,

A pedestrian wishing to cross the street at an uncontrolled intersection can only do so if its expected time slot for using the intersection does not interfere with that of an approaching vehicle.

Further:

In the simulation, pedestrians will only use such a crossing if the whole length of the crossing is free of vehicles for the whole time needed to cross.

Would it also be possible to add a pedestrian model that tries to interact with a driver/vehicle and, therefore, vehicles will stop to let the pedestrian cross? This could be based on some probability or, for autonomous vehicles, be ensured (probability = 1). This probability should be adjustable for autonomous vehicles.

In general, making the pedestrian model more customizable would be very nice. For example, research indicates that there are numerous factors influencing crossing decisions (see Rasouli and Tsotsos). These include distance, shape of the vehicle, velocity, and in the future, also probably the autonomy (see Millard-Ball and Ackermans et al.) as well as external communication of autonomous vehicles. With such flexibility, the impact of realistic pedestrian flow can be studied, even when autonomous vehicles are included.

Enabling this would make crossings at unsignalized streets in densely crowded cities, pedestrian areas, but also in general, more realistic (see Hong Kong or Singapore).

[namdre]

The vType attributes jmIngoreFoeProb and jmIgnoreFoeSpeed can already be used to make pedestrians cross at an unprioritized crossing despite oncoming vehicles. See https://sumo.dlr.de/docs/Definition_of_Vehicles,_Vehicle_Types,_and_Routes.html#junction_model_parameters

[M-Colley]

Interesting, thank you for the clarification.

Would it be possible to split jmIgnoreFoeProb for pedestrians, vehicles, and even autonomous vehicles as I am only interested in pedestrians ignoring rules? Additionally, it would be really great to make this dependent on whether a manually driven vehicle or an autonomous vehicle approaches.

Again, this need is based on the work by Millard-ball and Ackermans et al. that argue that pedestrian behavior will alter for autonomous vehicles compared to manually driven vehicles and there will, probably, be a longer period of mixed traffic.

[namdre]

jmIgnoreFoeProb is set per vType (persons also have a vType). This means different vehicles and persons can use different values.

[M-Colley]

Thanks, but this is probably then set for the pedestrian without accounting what kind of vehicle is approaching, right?

For example: Person A has the probability to cross of 0.8, then Person A will cross in front of any vehicle with the probability of 0.8. It would be awesome to be able to define that for a Vehicle (type) A, the probability is 0.8, and for Vehicle B only 0.1.

[namdre]

Yes. Setting a probability to step in front of a vehicle based on the type of that vehicle would be a novel feature.

[M-Colley]

Okay, will you consider implementing such a feature?

I understand that you will have a lot of other work to do, so I am just curious how the process would be here :)

[namdre]

I will consider implementing it but since it would be one of 1579 open issues you are better served by obtaining the desired functionality via TraCI: For each pedestrian, check the oncoming vehicles and then you compute whether to walk or stop (eventually setting jmIgnoreFoeProb=0 or =1)

[OrionZou]

use sumo1.60. When the vehcle enters the crosswalk, the pedestrian will stop waiting for the vehcle to pass. I don't want the pedestrian to stop. for vehcle and pedestrian, I set jmIgnoreFoeSpeed 100, a high speed, and set jmIngoreFoeProb 0 or 1.However, the pedestrian also stop for the vehcle to pass. what should I do？

[namdre]

There is currently no option that would make the pedestrians model walk into an obstacle by itself. The only way to get this behavior is to forcefully place the pedestrian in an obstacle using traci.person.moveToXY.

[OrionZou]

Thank you very much, I will try your suggestions. It needs to be explained is that ：when a car passes behind a pedestrian in an crosswalk, the pedestrian will still stop. Is this the same reason as you mentioned?

[namdre]

No. That sounds like a bug. The person should only stop walking while on the crossing if the car is actually in it's way. Please provide a minimal example to reproduce this.

[OrionZou]

File1: exp.sumocfg

File2: new_cross3.net.xml <?xml version="1.0" encoding="UTF-8"?>

File3: routes.rou.xml

File4: demo.py import traci from sumolib import checkBinary import time

if name == 'main': path = './exp.sumocfg' sumoBinary = checkBinary('sumo-gui') cmd = [sumoBinary, "-c", path] traci.start(cmd) ved_id = None while True: traci.simulationStep() if len(traci.vehicle.getIDList()) > 0: ved_id = traci.vehicle.getIDList()[0] traci.vehicle.setSpeedMode(ved_id, sm=0) traci.vehicle.setSpeed(ved_id, 6)

traci.vehicle.setLaneChangeMode(ved_id,lcm=0)

        traci.simulationStep()
        break
i = 0
while True:
    i += 1
    traci.simulationStep()
    time.sleep(0.1)
    if (i >= 78) and (i <= 82):
        print('pf1_.1 stopped')
    print(i)

[OrionZou]

File1's code view has problem, so I send a picture about it

[OrionZou]

In fact, I set traci.vehicle.setSpeedMode(self.curr_vehicle_id, sm=0), then when the vehcle enters the crossing and the pedestrian is very close to the vehcle, the pedestrian will stop, even if the pedestrian is in front of the car! the pedestrian is too stupid

[namdre]

You can zip your files and attach the zip. That would be much easier.

[namdre]

Note, I do not intend to re-type your file contents based on a screenshot. Please attach a zip file with the xml files.

[OrionZou]

Sorry, here is the .zip mini-sumo.zip

[OrionZou]

notice that path = './exp.sumocfg' in demo.py need to be insted of your file path.

[namdre]

I can confirm your bug: #7853 Some unrelated pointers for your example network:

• pedestrians can walk in both directions on the same edge. Thus you do not need back-and-forth pedestrian edges
• your pedestrian crossing is overlayed with a walkingArea because the roads on the right also allow pedestrian movements. For details, see https://sumo.dlr.de/docs/Simulation/Pedestrians.html#building_a_network_for_pedestrian_simulation
• you don't have to slow down the simulation via 'sleep' in your python script. Sumo supports the option --delay for setting delay directly in the gui.

[OrionZou]

Thank for your confirm ang your opinions. I am trying to use traci.person.moveToXY to achieve my requirement and I found:

I use traci.person.moveToXY. After a pedestrian have walked one edge, he cannot enter the next edge. Could you have a goo solution？ Is there a good way to judge that the pedestrian has completed the current edge and needs to move to the next edge？

[namdre]

Make sure that the pedestrian has a complete route. Moving it to an edge not on it's route will set a new route that consists of a single edge. You can check/change the route with getStage and replaceStage.

[OrionZou]

I think I set a complete route, but the pedestrian is stuck on the edge of the crosswalk and cannot complete the route. Could you please help me look at the code? Thank you very much! I re-uploaded the code due to a careless bug mini-sumo-v3.zip

[M-Colley]

I will consider implementing it but since it would be one of 1579 open issues you are better served by obtaining the desired functionality via TraCI: For each pedestrian, check the oncoming vehicles and then you compute whether to walk or stop (eventually setting jmIgnoreFoeProb=0 or =1)

Hey, so I've looked at the documentation and wanted to ask, regarding the implementation, whether there is a function to get all oncoming vehicles or whether I will have to get all vehicles and then check "manually" whether they are near the pedestrian? This could be a problem when they are, say, 30 m away but there is a parallel street or a junction with other vehicles.

[namdre]

Hey, so I've looked at the documentation and wanted to ask, regarding the implementation, whether there is a function to get all oncoming vehicles or whether I will have to get all vehicles and then check "manually" whether they are near the pedestrian? This could be a problem when they are, say, 30 m away but there is a parallel street or a junction with other vehicles.

you can call traci.lane.getFoes(person.getLaneID(personID), person.getNextEdge(personID) + "_0") to get the lanes that need to be checked for incoming vehicles. If that call returns nothing, reverse the arguments (happens if the person is walking against the natural direction of the edge).

[namdre]

I think I set a complete route, but the pedestrian is stuck on the edge of the crosswalk and cannot complete the route. Could you please help me look at the code? Thank you very much! I re-uploaded the code due to a careless bug mini-sumo-v3.zip

This was a bug: https://github.com/eclipse/sumo/issues/7907

[RickRocher]

I also think that the interaction between vehicle (per type - autonomous vs manual) and pedestrians should be more customizable. This would be a great feature and would help to plan pedestrian flow for the future.

Thanks for considering!

[namdre]

for reference, some of the desired functionality may be achieved via https://sumo.dlr.de/docs/Definition_of_Vehicles%2C_Vehicle_Types%2C_and_Routes.html#transient_parameters