how to make "CppNumericalSolvers" work

[xiaoyuanzh]

sry.. but can you help me how to compile CppNumericalSolvers?

[jsford]

Are you interested in using my code with CppNumericalSolvers, or just using their code? What build system are you using (if any)? Can you give more details? You might have more luck asking them directly over at https://github.com/PatWie/CppNumericalSolvers

[xiaoyuanzh]

Oh, yes. I have some troubles to compile your codes together with CppNumericalSolvers yesterday because of c++ version and 32/64 bit problems and also -PIC errors. But , luckily I fix them today. But, indeed, your write good codes, which is extremely helpful for me to design an agv on the road. A suggestion is that your can write cmakelist more machine/version-friendly next time(smile). Some questions are: 1. Do you think it is possible to write a trajectory tracker on your code> 2. How to deal with a trajectory with a bounded curvature?

[jsford]

Yeah I mostly just write my research code to run on my machine.

1. Sure! That's where I was headed. For example, you could use this code to do pure pursuit. (Generate a polytraj from your current pose to the pure pursuit lookahead pose).
2. You can use an optimization method that supports constraints on the variables being optimized. Check out the Al Kelly paper. Also look up Matthew McNaughton's thesis. He uses this method on a GPU to do trajectory planning.

[xiaoyuanzh]

Hi jsford! Questions about pure pursuit. When i have got a look ahead point. I use current position of robot as start state and look ahead point as end state. And what velocity should I send to my robot? Because if we use 100(or smaller) sample number of the spline, I can only get very small a starting velocity .