This repository contains the source code for our work on perceptive model predictive control. Please find the full text of the paper here.
J. Pankert and M. Hutter, “Perceptive Model Predictive Control for Continuous Mobile Manipulation,” IEEE Robotics and Automation Letters, vol. 5, no. 4, pp. 6177–6184, Oct. 2020.
@article{pankertPerceptiveModelPredictive2020a,
title = {Perceptive {{Model Predictive Control}} for {{Continuous Mobile Manipulation}}},
author = {Pankert, Johannes and Hutter, Marco},
year = {2020},
month = oct,
volume = {5},
pages = {6177--6184},
issn = {2377-3766},
doi = {10.1109/LRA.2020.3010721},
journal = {IEEE Robotics and Automation Letters},
number = {4}
}
Videos of our hardware experiments can be seen here.
The software has been tested with Ros Melodic under Ubuntu 18.04.
Create a new catkin workspace, configure it to build in release, download all dependencies with wstool and build perceptive_mpc.
sudo apt install python-catkin-tools libglpk-dev python-wstool -y
source /opt/ros/melodic/setup.bash
mkdir perceptive_mpc_ws
cd perceptive_mpc_ws
mkdir src
catkin init
catkin config --extend /opt/ros/melodic --cmake-args -DCMAKE_BUILD_TYPE=Release
cd src
git clone https://github.com/leggedrobotics/perceptive_mpc.git
wstool init . ./perceptive_mpc/perceptive_mpc_https.rosinstall
catkin build perceptive_mpc
Alternatively, you can also create a docker image by running:
docker image build -t perceptive_mpc:v0.3 .
or pull the image from dockerhub:
docker pull rslethz/perceptive_mpc
The easiest way to test the software is to use the provided launchfiles in the perceptive_mpc package.
roslaunch perceptive_mpc demo.launch:
This will launch a kinematic simulation of the motion planner. The computed optimal state is set as the observation of the MPC. An end-effector target can be set with an interactive marker in RVIZ.
If you use the docker image, run the following commands instead:
xhost local:root
docker container run -it --rm --name mpc_demo \
-e DISPLAY=$DISPLAY \
-v /tmp/.X11-unix:/tmp/.X11-unix \
--device /dev/dri \
perceptive_mpc:v0.3 ./src/perceptive_mpc/scripts/run_demo.sh
roslaunch perceptive_mpc collision_avoidance_demo.launch:
Launches the MPC controller with collision avoidance. Call the following service to load a Euclidian Signed Distance Field (ESDF) map:
rosservice call /voxblox_node/load_map "file_path: '/path/to/esdf/map.esdf'"
An demo map example_map.esdf is contained in the example directory.
If you use the docker image, run the following commands instead:
xhost local:root
docker container run -it --rm --name collision_avoidance_demo \
-e DISPLAY=$DISPLAY \
-v /tmp/.X11-unix:/tmp/.X11-unix \
--device /dev/dri \
perceptive_mpc:v0.3 ./src/perceptive_mpc/scripts/run_demo_collision_avoidance.sh
After everything started, load a map by running:
docker exec collision_avoidance_demo ./src/perceptive_mpc/scripts/load_map.sh
The ZMP mechanical stability constraint is active by default.
The radius of the support circle (circle inscribing the support polygon) can be set in the task.info configuration file.
In the kinematic simulation, the external wrench is set to a default value specified in the kinematic_simulation_parameters.yaml file. The wrench is specified in end-effector reference frame.
By sending a WrenchPoseTrajectory message, a time varying wrench can be specified. The wrenches in this message are again specified in the desired end-effector reference frame. The end-effector references themselves need to be specified in world frame.
The software can easily be integrated with a mobile manipulator platform. The configuration can be copied from perceptive_mpc/example/KinematicSimulation.cpp. The tracker thread is the main control loop. The current state estimate has to be set as the system observation. The optimal control inputs can be forwarded to the motor controllers.
The admittance control module can only be tested on hardware or in a physics based simulation. The source code of KinematicSimulation.[h|cpp] contains hints on how to interface with the module.
In order to use your custom manipulator, derive from the KinematicsInterface class and override the purely virtual methods for forward kinematics computation. We provide kinematic implementations for some popular robots:
- Mabi Speedy 12
- UR10
- UR5
- UR3
The kinematics code was generated with RobCoGen v0.4ad.0. Robcogen requires robot descriptions in the .kindsl/.dtdsl format. The urdf2robcogen tool converts to this format from .urdf files.
BSD-3-Clause Copyright (c) 2020 Johannes Pankert pankertj@ethz.ch All rights reserved.
Redistribution and use in source and binary forms, with or without modification, are permitted provided that the following conditions are met:
- Redistributions of source code must retain the above copyright notice, this list of conditions and the following disclaimer.
- Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution.
- Neither the name of this work nor the names of its contributors may be used to endorse or promote products derived from this software without specific prior written permission.
THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.