Better parallelization - Githubissues

pangtao22 commented 2 years ago

Includes an example on iRS-MPC for the allego hand, whose increased need for computation motivates a more careful parallelization implementation, including:

More modular API for parallelization. Specifically, most parallelization-related functions have been collected into a new class called QuasistaticDynamicsParallel.
- Cleaner ZMQ-based parallelization.
- Support for a new MPI-based parallelization implementation in C++, although it is 30-40% less performant on my 2018 Mac mini. Could be faster on ubuntu?
Renaming LQR to MPC.

This change is

pangtao22 commented 2 years ago

I think it's ready for experiments and feedback from users! The dash apps are not up-to-date yet, but running iRS-MPC on both examples should work!

hjsuh94 commented 2 years ago

How should we handle the sequence of PRs?

I think we should review / merge this one first and then worry about how merging / refactoring #1 ?

pangtao22 commented 2 years ago

Agreed! This PR only refactors iRS-MPC and adds Zmq-free parallelization. I hope this provides a more solid starting point for future PRs on planning.

pangtao22 / planning_through_contact

Better parallelization #5