Issues with Implementing the Controller on Real Hardware

[EthanZengg]

Thank you for open-sourcing the controller. I might encounter some simple issues and I hope to get your reply.

Q1
I'm encountering slower MPC iteration speeds than those reported in the paper when running the example on my system, despite utilizing parallel computing. My CPU is Intel Core i9-14900k with 24 cores and 32 threads and my system achieves 40.27 ms per iteration for kino-dynamics modle, whereas the paper reports 20.3 ms.

Q2
In the paper "From centroidal to whole-body models for legged locomotion: a comparative analysis," the MPC is designed to run at a frequency of 100 Hz . However, achieving this frequency is challenging for the kino-dynamics model. if I want to use this controller on Real Hardware, How should I adjust the control delay parameter which was set to 10 in the simulation ?

for j in range(10): q_meas, v_meas = device.measureState() x_measured = np.concatenate([q_meas, v_meas])

[edantec]

Hi Ethan,

Sorry for the delayed answer.

Make sure you are compiling simple-mpc with this branch of Aligator: https://github.com/edantec/aligator/tree/temporary_fix

The additional computation load may result from an unnecessary memory reallocation happening at each control cycle. We are working on a new memory allocation scheme that will address this issue when using the parallel solver.