Collision avoidance MPC in acados - Dynamic obstacles and signed distance fields

MPC Discussions
1

HI all,

I want to implement a MPC with collision avoidance for kynodynamic planning of an UAV. This UAV has a lidar onboard. I wondered what the possibilities are to include this lidar data in my MPC. Currently I’m still experimenting in simulation where I can use a simple constraint such as h(x)>=0, where h(x) is the distance to a fixed static object.

My questions are as follows:

  • How would this change if h(x) changes over time? I.e., for dynamic obstacles.
  • Can I add a voxblox signed distance field in acados? I.e. h(x)>=0 where h(x) is the signed distance field.
  • Is there another more obvious strategy to do this kynodynamic planning?

I’m far from an expert in this area so any help is welcome! Thanks :slight_smile:

2

Hi,

you could introduce parameters and update them to be the latest measurements of your obstacles.

Since acados implements SQP-type methods, it is essential to have smooth functions. Voxblox seems to be something discrete by definition and thus not really suitable.

Good luck with your implementation!