search

chraibi / sfm-oscillation

http://arxiv.org/abs/1412.1133
1 stars 1 forks source link

About sfm-oscillation and jam #1

Open godisreal opened 5 years ago

godisreal commented 5 years ago

I think your repo jam is a upgrade version of sfm-oscillation. Is that right?
I take a look at your python code of sfm-oscillation, and it is good and consice. Appreciate.
Below is a simulation result by running your code.
xnResult2

Comments: (1) You may consider oscillation is a problem of social force model. This makes sense from the perspective of physics and engineering (ODE). But I have a diiferent perspective to understand this issue. Would you like to hear about my viewpoint?
(2) In your manuscript in arXiv, the model of the python code is well presented. An assumption of your model is r_ij=0. This implies that you omit the physical size of agents. Is that right? Actually it is also OK to assume r_ij to be a constant, and there is no major difference in the oscillation result. The above figure is obtained by assuming r_ij = constant (e.g., 0.6).
(3) There is also a PDE analysis of social force model, and I think oscillation also exists in PDE analysis. What do you think?

chraibi commented 5 years ago

I think your repo jam is a upgrade version of sfm-oscillation. Is that right?

I'm not sure what you mean with upgrade, but the jam repository is an implementation of a different model described in this article. See also the calculation of the repulsive force here f_rep = -v0 / tau * np.log(c * R + 1))

and compare it to this line in this repository f_rep = -a*np.exp(-dist/b)

I take a look at your python code of sfm-oscillation, and it is good and consice. Appreciate.

Thanks!

Comments: (1) You may consider oscillation is a problem of social force model. This makes sense from the perspective of physics and engineering (ODE). But I have a diiferent perspective to understand this issue. Would you like to hear about my viewpoint?

Sure! Please explain your perspective.

(2) In your manuscript in arXiv, the model of the python code is well presented. An assumption of your model is r_ij=0. This implies that you omit the physical size of agents. Is that right? Actually it is also OK to assume r_ij to be a constant, and there is no major difference in the oscillation result. The above figure is obtained by assuming r_ij = constant (e.g., 0.6).

You are right. The (constant) size of pedestrians is not relevant in this analysis I think.

(3) There is also a PDE analysis of social force model, and I think oscillation also exists in PDE analysis. What do you think?

Would you please show me a source for this PDE analysis?

godisreal commented 5 years ago

In brief, I change your code slightly as below.

f_rep = -a*np.exp(d0-dist/b)

d0 is the desired interpersonal distance, and it is a counterpart of desired velocity v0 in the self-driving force. It is not a constant, and it may vary as the environment changes. So oscillation may exist in d0, not in the actual distance dist. In other words, it is desired distance that oscillates rather than the physical distance.

We commonly do not observe oscillating walkers in the realistic life, but the simulation shows possibility of oscillation. So the simulation justifies that the social force model should be modified or improved in some aspects. Here a key issue is that we model the human crowd, not particles in traditional physics.

Not sure how to present the PDE analysis. In brief you may consider it as the fluid-based analysis of social force model, or as Payne-Whitham traffic flow model with social force applied. The resulting equation is PDE. I suppose oscillation can also be derived there.

godisreal commented 5 years ago

I also tested oscillation of group social force that I used in my simulation. Group social force is either attractive or repulsive. The result shows that the group social force may have intensive oscillation when people gets to the equillibrium position. So I combine it with v_ij component in GCFM to offset the oscillation. The result looks good. I will explain the result in a brief document and share it with you in the near future.

chraibi commented 5 years ago

Thanks for sharing. Please drop me an E-Mail. I would like to read your document.

godisreal commented 4 years ago

Happy Thanksgiving! I sent you a document two weeks ago. Have you got it? Your comment is much welcome.
Also, I read your manuscript on arXiv.org. I think the model should be linearized at the equillibrium position, but the equillibrium position is not at Delta x =0 (i.e., x_n+1 - x_n = 0). Please check.
In the existing model the equillibrium position is not explicitly shown in the equation, and it is a function of v_0 by calculation. I think this is why you add a v_0 term in GCFM to tune the equillibrium position.

chraibi commented 4 years ago

please send me your Email at m.chraibi@gmail.com

I did not find your document.

godisreal commented 4 years ago

Hi, Chraibi, I sent you an email on Nov. 17, 2019 from China. I thought it was delivered. I attached a document to my email and maybe it is filtered out by your email rule. Never mind. I have uploaded the document on my repo group-social-force/doc/oscWalker2019.pdf and you can directly download it there. Your comments are much welcome if you are interested.
Sorry for late reply.