Is there any effective algorithm for collision avoidance of moving cars in a street, by only using decrease or increase speed?

I have read some other algorithms such as "steering collision avoidance" which is for fixed obstacles. and they only change the path and speed of the entity to avoid collision, but what if entities are moving and they are not able to change the heading? for instance assume that two cares are moving in two crossing streets, and the only thing that they can do to avoid collision is to decrease their speed.

I was in though that I can use OBB's and check whether the two OBB's might collide in future or not. is there any better solution?

I don't think you can do much better, and there are situations you cannot get out of. Eg two cars in a straight line pointing towards each other, are going to collide eventually, unless you can reduce to 0 speed (which gives you a deadlock), or you can reduce to negative speeds (which would work, but may not be valid).

In real-life, we tend to use traffic lights for these occasions, giving exclusive access to the shared piece of the street to one of the cars. Without traffic lights, we run a computation in our head whether we would collide, and if so, reduce speed, or stop. Traffic rules or mutual agreement then avoid both sides doing the same computation, and both slowing down, until you're in a deadlock again.

It goes beyond cars, too. We all do this. If people are walking along converging paths (e.g. 90­­­° angle), we actually change speeds more often than we change direction. Decent method would be predicting if you are going to collide by some sort of look-ahead. If so, then the one who is furthest from the projected collision point slows down. However, it is possible that both would slow down -- which is a realistic looking result. Then just choose randomly who resumes normal speed and the other passes behind.

This is a long-standing set of research in various fields. The umbrella term is typically "Velocity Obstacles".

A lot of algorithms exist but most are minor variants on a theme. It mostly boils down to what edge cases you're willing to tolerate breaking down (i.e. "looking bad").

This triggered a memory I had of this article, which seems relevant:

https://arstechnica.com/science/2015/01/collision-avoidance-predicts-pedestrians-behavior/

Worth watching, btw.

http://www.gdcvault.com/play/1021986/Forced-Based-Anticipatory-Collision-Avoidance

This is a long-standing set of research in various fields. The umbrella term is typically "Velocity Obstacles".

A lot of algorithms exist but most are minor variants on a theme. It mostly boils down to what edge cases you're willing to tolerate breaking down (i.e. "looking bad").

Thank you, this was quite useful to me, I have read the original paper ("Motion Planning in Dynamic Environments using Velocity Obstacles [Fiorini 1998]"). Yet I have some problems in implementing it.

is there any implementation of the algorithm in C++? and if not, what is "configurationSpace" and "Minskowski sum vector", how can I implement them in C++?