rattle-iros-2022

Parametric information-aware motion planning using the RATTLE method, as published in IROS/RA-L 2022.

What is this?

RATTLE is a set of modules that perform parametric information-aware motion planning; in other words, RATTLE creates motion plans for robots to follow that try to be both useful for learning about system unknowns while also attempting to reach a goal/minimize fuel use. Uniquely, RATTLE includes capabilities to use learned system information on-the-fly, allowing the robot to better understand its own system model, while reasoning at a control level about how cautious it should be. Please see the paper for more detail.

RATTLE is implemented here for a 6DOF free-flying robot, though its interfaces and methods are more general to robotic systems with parameteric unknowns in their system dynamics. Specifically, RATTLE has been tested on the International Space Station using the Astrobee robots. A detailed account of code deployment and results of the flight experiments is given in this recently submitted paper.

RATTLE running and replanning in a simulation environment (top), and demonstrating a trajectory with weighted information content on the ISS (bottom).

Requirements

For coordinated use of all packages with the Astrobee sim Ubuntu 16.04 or 20.04 is required (possibly 22.04 depending on Astrobee simulation support) with a standard ROS installation.

• CasADi 3.5.5

git clone https://github.com/casadi/casadi --branch 3.5.5
cd casadi && mkdir build && cd build  # this is the top-level directory
# GOTO line 614, and `set(WITH_LAPACK ON)`
# GOTO line 626, and `set(WITH_QPOASES ON)`
# (if you can find a cleaner way of specifying this please let us know!)
cmake ..
make -j2
make install  # You might need sudo

• flann

apt install libflann-dev

• Misc. Python Dependencies

pip3 install matplotlib scipy numpy pycddlib

• gmp-dev

apt install libgmp-dev

• ACADO (bundled)

• Pytope (bundled)

• Bullet (bundled)

• Astrobee custom msgs and classes (bundled)

• Luajit (likley already on your system)

apt install libluajit-5.1-dev

• Autograd

pip3 install autograd

• pycddlib

pip3 install pycddlib

Installation

Dependencies

See above.

RATTLE packages

Create a ROS workspace and set up the RATTLE packages within it:

mkdir rattle-ws
cd rattle-ws
catkin init
git clone https://github.com/albee/rattle-iros-2022
mv rattle-iros-2022 src/rattle

Astrobee Simulation Setup

This implementation of RATTLE features tight integration with the Astrobee simulation environment. Get a compatible version:

export ASTROBEE_WS=${HOME}/rattle-ws/
git clone https://github.com/nasa/astrobee.git $ASTROBEE_WS/src/astrobee
pushd $ASTROBEE_WS/src/astrobee
git checkout v0.16.1
git submodule update --init --depth 1 description/media
popd

Now, build Astrobee's dependencies:

sudo apt update
sudo apt upgrade

pushd $ASTROBEE_WS
cd src/astrobee/scripts/setup
./add_ros_repository.sh
sudo apt-get update
cd debians
./build_install_debians.sh
cd ../
./install_desktop_packages.sh
sudo rosdep init
rosdep update
popd

Finally, configure and run catkin to build both the Astrobee sim and TRACE:

pushd $ASTROBEE_WS
./src/astrobee/scripts/configure.sh -l -F -D

# This step is very important! If your paths are wrong, you will not be able to build Astrobee packages.
export CMAKE_PREFIX_PATH="${CMAKE_PREFIX_PATH:+"$CMAKE_PREFIX_PATH:"}${ASTROBEE_WS}/src/astrobee/cmake"
catkin build -j2

Don't forget to source your workspace!

source $ASTROBEE_WS/devel/setup.bash

This will build the RATTLE packages, which can be used standalone, or coordinated as a whole with the aid of the Astrobee simulation environment (see below).

Usage

RATTLE's packages can be used standalone, or as a coordinated whole as in the examples in the paper. To run as a whole, RATTLE requires a simulation environment to respond to and provide message inputs/outputs, which you should have installed above.

ISS-Like Commanding Environment

You can launch RATTLE in the Astrobee simulation environment using rattle_astrobee_interface and its execute_asap interface:

• Launch space environment (ISS):

roslaunch rattle_astrobee_interface sim_rattle.launch world:=iss rviz:=true

• Launch ground environment (ground):

roslaunch rattle_astrobee_interface sim_rattle.launch world:=granite rviz:=true

• When switching between environments be sure to reset accelerometer bias:

rosrun executive teleop_tool -ns "queen/" -reset_bias

• Run a RATTLE test (after sim launch)

`rosrun execute_asap pub_gds_topics.py --sim 0` : to run test 0 on the ISS for sim.

`rosrun execute_asap pub_gds_topics.py --ground --sim 0` : to run test 0 on the ground for sim.

`rosrun execute_asap pub_gds_topics.py --sim -1` : to stop a test. You will see lots of output as the nodes are stopped.


`[-g, --ground]` : Run a ground test. Default to ISS.

`[-s, --sim]` : Run a simulation test. Defaults to hardware.

Tests19 and 78 are good demonstration tests to try out next---see the full list below!

• See execute_asap/README.md for high-level usage instructions to understand how nodes are started/stopped and the execution flow of test commands.

• Note that tests start from desired initial conditions! You might need to rerun a test to allow Astrobee to reach the proper start state.

Test List

• -1 Stop and reset nodelets
• 0 Quick checkout of RATTLE software
• 9 Non-info aware trajectory, from point B to C
• 10 Mass info-gain only, B to C
• 11 Inertia info-gain only, B to C
• 12 Info gain for all 4 parameters (mass and inertias)
• 13 Tube MPC checkout test ("MIT")
• 15 Standard MPC checkout test ("MIT")
• 16 RATTLE test: generic obstacle, run full pipeline (no est updates used)
• 17 RATTLE test: "hard" obstacle, run full pipeline
• 18 RATTLE test: run acado planner only
• 19 RATTLE test: run with high weight, then low weight
• 20 RATTLE test: gather translation data for estimation checkout
• 21 RATTLE test: mass excitation
• 22 RATTLE test: I_zz (moment of inertia) excitation
• 77 RATTLE full test (A to B)
• 78 RATTLE replan test (avoid an "astronaut" obstacle!)

Visualization

Astrobee's simulation environment provides visualization using RViz. To visualize RATTLE topics, a custom RViz file is available in rattle_astrobee_interface/config/iss_rattle.rviz. To open this manually in RViz, select File > Open Config.

Standalone Usage

Most packages separate ROS wrappers from core algorithms for standalone use; please consult individual READMEs in each package to adapt for use with other simulation frameworks.

ROS Packages

RATTLE's functions are implemented here as separate ROS packages:

• Global planning

  • rattle_rrt: A "long-horizon" sampling-based planner that also accounts for, in this case, translational dynamics.

• Local information-aware planning

  • rattle_acado_planner: A mid-level "local" planner that also considers information content of planned trajectories. Implemented using ACADO.

• Low-level robust control

  • casadi_nmpc : A robust tube MPC control implementation, for the translational dynamics.
  • z_poly_calc: An mRPI (minimum robust postiviely invariant set) calculator, using Rakovic's method.

• Online parameter estimation

  • inv_fam: Astrobee's Force Allocation Module (or FAM) computes the nozzle opening angles required to apply commanded wrenches. The inv_fam package estimates the true (post-saturation) forces and torques from nozzle openings, for use by the inertial parameter estimator
  • param_est: A sequential inertial parameter estimator - this implementation considers rigid body dynamics and estimates the mass and principal moments of inertia.

• ROS

  • rattle_msgs: Custom msg types used by these packages.
  • rattle_coordinator: The coordinator node that runs eveything and oversees execution. Uses a messy version of ASAP, which has shiny cleaned up interfaces now.
  • execute_asap: A manager node that launches all other nodes. Useful if hardware testing, especially using Astrobee.
  • data: Miscellaneous I/O.

Citation

If you find RATTLE is useful in your own work, please consider citing our relevant papers:

@article{albeeRATTLEMotionPlanning2022,
  title = {The RATTLE Motion Planning Algorithm for Robust Online Parametric Model Improvement with On-Orbit Validation},
  author = {Albee, Keenan and Ekal, Monica and Coltin, Brian and Ventura, Rodrigo and Linares, Richard and Miller, David W.},
  year = {2022},
  journal = {Robotics and Automation, Letters}
}

@article{doerrReSWARMMicrogravityFlight2024,
  title = {The ReSWARM Microgravity Flight Experiments: Planning, Control, and Model Estimation for On-orbit Close Proximity Operations},
  shorttitle = {The ReSWARM Microgravity Flight Experiments},
  author = {Doerr, Bryce and Albee, Keenan and Ekal, Monica and Ventura, Rodrigo and Linares, Richard},
  year = {2024},
  journal = {Journal of Field Robotics}
}