How about adding a new circle configuration to MPT

[mehmetdogru]

Currently for MPT algorithm we have 3 circle configs. If following parameters are set to 3 related configs are as follows:

• uniform_circle.num
• rear_drive.num_for_calculation
• bicycle_model.num_for_calculation

⚠️ All of these configurations have lateral overhangs.

If these lateral overhangs are unwanted in such cases where the ego vehicle should drive on narrow roads, only way to do that to increase the parameters above for selected configuration. In that case the radius of circles converges to vehicle_width / 2 yet of course can never be equal to vehicle_width / 2. Also usage of greater value for circles/calculation numbers to converge the radius of circles to vehicle_width / 2, highly increases the calculation cost.

const double radius =
    std::hypot(vehicle_param.length / static_cast<double>(circle_num) / 2.0, vehicle_param.width / 2.0) * radius_ratio;

---

So my proposal is to add a new circle configuration for MPT which does not have lateral overhangs. This configuration needs at least 2 circles at the front and the back of the vehicle. And the rest of the circles (fitting_uniform_circle.num - 2) are simply interpolated along the vehicle length.

What do you think about this idea?

cc: @TakaHoribe @takayuki5168 @mitsudome-r @xmfcx

[takayuki5168]

@mehmetdogru Thank you for your suggestion.

In my understanding, each method has pros/cons. For example, in this case with a narrow and tight curve road, the original one is a bit better.

But we cannot estimate the behavior of optimization when changing the condition, so it's fine for me to add the implementation of your idea.

[takayuki5168]

I feel the interval of the circle's center points should be not vehicle_length / n but vehicle_length / (n - 1).

[mehmetdogru]

But we cannot estimate the behavior of optimization when changing the condition, so it's fine for me to add the implementation of your idea.

Thank you for your input. Yes, as you said there are pros/cons related to this topic. Then I will create a PR about this. 😃

I feel the interval of the circle's center points should be not vehicle_length / n but vehicle_length / (n - 1).

Sorry, my mistake! Thank you.

[takayuki5168]

@mehmetdogru This discussion can be resolved?

[mehmetdogru]

Sorry for the late reply, it certainly can.

[takayuki5168]

@mehmetdogru
Thank you for creating the PR! I'm curious about which method is better.
If you test and compare the original one and the new one, please let me know the result.

[mehmetdogru]

Thank you. Sure will do!