Custom Task Interactive#

The following example solves a qp problem expressed at the joint velocity level such that:

\[\begin{split}\begin{array}{ccc}\boldsymbol{\dot{q}}^{opt} = & \underset{\boldsymbol{\dot{q}}}{\mathrm{argmin}} & ||J(\boldsymbol{q})\boldsymbol{\dot{q}} - \boldsymbol{v}^{target} || + \omega || \boldsymbol{\dot{q}} ||^2\\& \textrm{s.t.} & \boldsymbol{\dot{q}^{min}} \leq \boldsymbol{\dot{q}} \leq \boldsymbol{\dot{q}^{max}}. \\ & & \boldsymbol{q}^{min} \leq \boldsymbol{q}(\boldsymbol{\dot{q}}) \leq \boldsymbol{q}^{max} \end{array}\end{split}\]

The robot main tasks is defined as a custom task and consists in following a simple trajectory defined in Cartesian space. The mujoco library is used to simulate the robot behaviour.

C++

Full Code:

You can find the source code of this example here.

// Copyright 2021 DeepMind Technologies Limited
//
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
//
//     http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.

#include "mujoco/mujoco_sim.h"
#include "Qontrol/Qontrol.hpp"
#include "trajectory_generation/trajectory_generation.h"

#define MUJOCO_PLUGIN_DIR "mujoco_plugin"


using namespace Qontrol;

class MujocoQontrol : public MujocoSim 
{
public:
//------------------------------------------- simulation -------------------------------------------
std::shared_ptr<Qontrol::Model::RobotModel> model;
std::shared_ptr<Qontrol::JointVelocityProblem> velocity_problem;
std::shared_ptr<Qontrol::Task::GenericTask> custom_task;
pinocchio::SE3 init_pose;

Qontrol::RobotState robot_state;

TrajectoryGeneration* traj;
std::string resource_path;
void initController() override
{
  model =
      Model::RobotModel::loadModelFromFile(resource_path+"robot.urdf");
  
  const int ndof = model->getNrOfDegreesOfFreedom();
  
  velocity_problem = std::make_shared<Qontrol::JointVelocityProblem>(model);  

  custom_task = velocity_problem->task_set->add("CustomMainTask", 6, 1.0);
  auto regularisation_task = velocity_problem->task_set->add<Task::JointVelocity>("RegularisationTask",1e-5); // <-- Based on the template given will implement correct task representationn

  auto joint_configuration_constraint = velocity_problem->constraint_set->add<Constraint::JointConfiguration>("JointConfigurationConstraint");
  auto joint_velocity_constraint = velocity_problem->constraint_set->add<Constraint::JointVelocity>("JointVelocityConstraint");

  mju_copy(d->qpos, m->key_qpos, m->nu);
  robot_state.joint_position.resize(ndof);
  robot_state.joint_velocity.resize(ndof);
  for (int i=0; i<ndof ; ++i)
  {
    robot_state.joint_position[i] = d->qpos[i];
    robot_state.joint_velocity[i] = d->qvel[i];
  }
  model->setRobotState(robot_state);
  
  traj = new TrajectoryGeneration(resource_path+"trajectory.csv", m->opt.timestep);
  
}

void updateController() override
{
  const int ndof = model->getNrOfDegreesOfFreedom();

  for (int i=0; i<ndof ; ++i)
  {
    robot_state.joint_position[i] = d->qpos[i];
    robot_state.joint_velocity[i] = d->qvel[i];
  }
  model->setRobotState(robot_state);

  pinocchio::SE3 current_pose(model->getFramePose(model->getTipFrameName()).matrix());
  traj->update();
  pinocchio::SE3 traj_pose(traj->pose.matrix());
  const pinocchio::SE3 tipMdes = current_pose.actInv(traj_pose);
  auto err = pinocchio::log6(tipMdes).toVector();
  Eigen::Matrix<double,6,1> p_gains;
  p_gains << 10,10,10,10,10,10;
  Eigen::Matrix<double,6,1> xd_star = p_gains.cwiseProduct(err);
  custom_task->setE(model->getJacobian(model->getTipFrameName()));
  custom_task->setf(xd_star);
  velocity_problem->update(m->opt.timestep);

  if (velocity_problem->solutionFound())
  {
    sendJointVelocity(velocity_problem->getJointVelocityCommand());    
  }
}
 
};

int main(int argc, const char** argv) {
  MujocoQontrol mujoco_qontrol;
  Qontrol::Log::Logger::parseArgv(argc, argv);

  mjvScene scn;
  mjv_defaultScene(&scn);

  mjvCamera cam;
  mjv_defaultCamera(&cam);

  mjvOption opt;
  mjv_defaultOption(&opt);

  mjvPerturb pert;
  mjv_defaultPerturb(&pert);

  // simulate object encapsulates the UI
  auto sim = std::make_unique<mj::Simulate>(
      std::make_unique<mj::GlfwAdapter>(), &cam, &opt, &pert, /* fully_managed = */ true
  );

  std::string robot = argv[1];
  std::string mujoco_model = "./resources/"+robot+"/scene.xml";
  mujoco_qontrol.resource_path = "./resources/"+robot+"/";

  // start physics thread
  std::thread physicsthreadhandle( &MujocoQontrol::PhysicsThread, mujoco_qontrol, sim.get(), mujoco_model);

  // start simulation UI loop (blocking call)
  sim->RenderLoop();
  physicsthreadhandle.join();

  return 0;
}

—

Explanation of the code

Declaration

First we declare all the objects that will be used to define our problem.

std::shared_ptr<Qontrol::Task::GenericTask> custom_task;

We use pinocchio for our model library.

  velocity_problem = std::make_shared<Qontrol::JointVelocityProblem>(model);  

The output of our qp controller is at the velocity level.

  custom_task = velocity_problem->task_set->add("CustomMainTask", 6, 1.0);

Here we declare the custom type as a GenericTask.

Initialization

void initController() override
{
  model =
      Model::RobotModel::loadModelFromFile(resource_path+"robot.urdf");

During initialization we instantiate the model with the robot urdf.

std::shared_ptr<Qontrol::JointVelocityProblem> velocity_problem;

We initialize the problem by giving it the model. By default, the qpmad library is used.

  custom_task = velocity_problem->task_set->add("CustomMainTask", 6, 1.0);

We fill the task set of velocity_problem with the custom task. We give it a name, the dimension of the task, and the relative weight relatively to the other tasks. Here the task is in Cartesian space so it uses 6 degrees of freedom. Since it is the main task it has a maximal priority relatively to the other tasks so we give it a weight of 1.

  auto regularisation_task = velocity_problem->task_set->add<Task::JointVelocity>("RegularisationTask",1e-5); // <-- Based on the template given will implement correct task representationn

We then fill the task set with the regularisation task. In this example, the regularisation tasks is defined as a joint veloicty task. Its means that this task will minimize the overall robot joint veloicty. It is given a small weight so that it doesn’t interfere with the main task

  auto joint_configuration_constraint = velocity_problem->constraint_set->add<Constraint::JointConfiguration>("JointConfigurationConstraint");
  auto joint_velocity_constraint = velocity_problem->constraint_set->add<Constraint::JointVelocity>("JointVelocityConstraint");

We then fill the constraint set of velocity_problem with the two pre-implemented constraints. Each constraint is given a name. These constraints will automatically be updated during the update of Qontrol.

  mju_copy(d->qpos, m->key_qpos, m->nu);
  robot_state.joint_position.resize(ndof);
  robot_state.joint_velocity.resize(ndof);
  for (int i=0; i<ndof ; ++i)
  {
    robot_state.joint_position[i] = d->qpos[i];
    robot_state.joint_velocity[i] = d->qvel[i];
  }
  model->setRobotState(robot_state);

We create the robot state and fill it with the simulated robot current state.

  traj = new TrajectoryGeneration(resource_path+"trajectory.csv", m->opt.timestep);

We create a simple trajectory that has been precalculated and store in a csv file. This trajectory start at the robot current Cartesian pose and does a translation of (-0.1, -0,1, -0.1) m.

Update

void updateController() override
{
  const int ndof = model->getNrOfDegreesOfFreedom();

  for (int i=0; i<ndof ; ++i)
  {
    robot_state.joint_position[i] = d->qpos[i];
    robot_state.joint_velocity[i] = d->qvel[i];
  }
  model->setRobotState(robot_state);

  pinocchio::SE3 current_pose(model->getFramePose(model->getTipFrameName()).matrix());
  traj->update();

The update function is called every milliseconds. At the beginning of each update we fill the new robot state according to the simulated robot.

We also update the trajectory so that it gives the next Cartesian pose to reach in 1 ms.

  pinocchio::SE3 current_pose(model->getFramePose(model->getTipFrameName()).matrix());
  traj->update();
  pinocchio::SE3 traj_pose(traj->pose.matrix());
  const pinocchio::SE3 tipMdes = current_pose.actInv(traj_pose);
  auto err = pinocchio::log6(tipMdes).toVector();
  Eigen::Matrix<double,6,1> p_gains;
  p_gains << 10,10,10,10,10,10;
  Eigen::Matrix<double,6,1> xd_star = p_gains.cwiseProduct(err);

We then compute the desired Cartesian velocity using a simple proportionnal controller. Pinocchio is used to compute the error between the desired Cartesian pose and the current Cartesian pose. This is done by the log6 function. The p_gains are the proportionnal gains of the controller.

  custom_task->setE(model->getJacobian(model->getTipFrameName()));
  custom_task->setf(xd_star);

We then update the terms of the custom task. The E matrix is equal to the current robot Jacobian matrix expressed at the tip of the robot. The f term is the desired Cartesian velocity computed previously.

  velocity_problem->update(m->opt.timestep);

  if (velocity_problem->solutionFound())
  {
    sendJointVelocity(velocity_problem->getJointVelocityCommand());    
  }
}
 
};

Once we updated the necassary tasks and constraints we can update the whole problem. If a solution to the problem exist we can then get it and send it to the simulated robot.

Main function

int main(int argc, const char** argv) {
  MujocoQontrol mujoco_qontrol;
  Qontrol::Log::Logger::parseArgv(argc, argv);

  mjvScene scn;
  mjv_defaultScene(&scn);

  mjvCamera cam;
  mjv_defaultCamera(&cam);

  mjvOption opt;
  mjv_defaultOption(&opt);

  mjvPerturb pert;
  mjv_defaultPerturb(&pert);

  // simulate object encapsulates the UI
  auto sim = std::make_unique<mj::Simulate>(
      std::make_unique<mj::GlfwAdapter>(), &cam, &opt, &pert, /* fully_managed = */ true
  );

  std::string robot = argv[1];
  std::string mujoco_model = "./resources/"+robot+"/scene.xml";
  mujoco_qontrol.resource_path = "./resources/"+robot+"/";

  // start physics thread
  std::thread physicsthreadhandle( &MujocoQontrol::PhysicsThread, mujoco_qontrol, sim.get(), mujoco_model);

  // start simulation UI loop (blocking call)
  sim->RenderLoop();
  physicsthreadhandle.join();

  return 0;

The main function function fetches the robot name given in argv and starts the Mujoco simulation.

Python