Skip to content

MeMory-of-MOtion/sobec

Repository files navigation

Sobec: Sandbox for optimal control explicitly for bipeds

Pipeline status Coverage report Code style: black

Sandbox for optimal control explicitly for bipeds

Minimal python version: 3.6

Current state of the MPC, Classes:

RobotDesigner

it is a robot wrapper (After, it will be called RobotWrapper)

ModelMaker

This class produces a std::vector of AbstractModelAction, it is done using the method formulateHorizon.

There is only one formulation implemented right now, specified on the method formulateStepTracker, it would be good to incorporate the other formulations in new methods such as formulateStairClimber or formulateWithoutThinking. Reuse the contacts and costs methods that are already made (with names starting by define...).

Once new formulations are made, it would be good to have a name based selector in the method formulateHorizon.

HorizonManager

it is the OCP (After, it will be called OCP)

It receives a vector of AbstractModelActions (created by the ModelMaker) and provides methods to deal with the ddp object of crocoddyl.

WBC

it is the MPC (After it will be called MPC)

It provides the method iterate that receives the measured state and returns the joint torques that should be commanded in the robot. All previous classes are used here.

MainControlLoop

It is missing, this script should instantiate the WBC and computes the control in a loop with ros.