FIGAROH-Plus is a fork of FIGAROH (by the same author), developed at LAAS-CNRS as the output of my thesis.

This toolbox provides a conprehensive framework for parametric system identification of multi-body robotic systems based on the popular modeling convention URDF. Currently, it only support tree-like kinematic structures including floatbase robots such as humanoids.

WIP:

• online identification
• mobile robot
• GUI
• parallel mechanism

1. Install python3-pip
2. Set up the robotpkg repositories in your source list as explained here: http:https://robotpkg.openrobots.org/debian.html Then, replace 3* with your python 3 version and then execute:

sudo apt-get install robotpkg-py3*-pinocchio robotpkg-py3*-hpp-fcl robotpkg-py3*-ndcurves

3. Install other dependencies

pip3 install --user numdifftools quadprog numpy scipy picos pandas meshcat pyyaml

4. Install cmake as explained here: https://apt.kitware.com
5. Install FIGAROH from source

git clone https://gitlab.laas.fr/gepetto/figaroh.git
git checkout -b pal-tiago-calib origin/pal-tiago-calib
git submodule init
git submodule update
mkdir build
cd build
cmake ..
sudo make install

Update python path, or put it permanent in ~/.bashrc

export PYTHONPATH=:/usr/local/lib/python3/dist-packages:$PYTHONPATH

As described in the following figure it provides:

• Dynamic Identification:
  • Dynamic model including effects of frictions, actuator inertia and joint torque offset.
  • Generation of continuous optimal exciting trajectories that can be played onto the robot.
  • Guide line on data filtering/pre-processing.
  • Identification pipeline with a selection of dynamic parameter estimation algorithms.
  • Calculation of physically consistent standard inertial parameters that can be updated in a URDF file.
• Geometric Calibration:
  • Calibration model with full-set kinematic parameters.
  • Generation of optimal calibration postures based on combinatorial optimization.
  • Calibration pipeline with customized kinematic chains and different selection of external sensoring methods (eye-hand camera, motion capture) or non-external methods (planar constraints).
  • Calculatation of kinematic parameters that can be updated in URDF model.

Overall, a calibration/identification project folder would like this:

\considered-system
    \config
        considered-system.yaml
    \data
        data.csv
    optimal_config.py
    optimal_trajectory.py
    calibration.py
    identification.py
    update_model.py

A step-by-step procedure is presented as follow.

• Step 1: Define a config file with sample template.
  A .yaml file containing information of the considered system and characteristics of the calibration/identification problem has a structure as follow:

  calibration:
      calib_level: full_params
      non_geom: False
      base_frame: universe
      tool_frame: wrist_3_link
      markers:
          -   ref_joint: wrist_3_joint
              measure: [True, True, True, True, True, True]
      free_flyer: False
      nb_sample: 29
  

  identification:
      robot_params:
          -   q_lim_def: 1.57 
              dq_lim_def : 5.0 
              ddq_lim_def : 20.0
              tau_lim_def : 4.0
              fv : None
              fs : None
              Ia : None
              offset : None
              Iam6 : None
              fvm6 : None
              fsm6 : None
              N : None
              ratio_essential : None
      problem_params:
          -   is_external_wrench : False
              is_joint_torques : True
              force_torque : ['All']
              external_wrench_offsets : False
              has_friction : False
              has_joint_offset : False
              has_actuator_inertia : False
              is_static_regressor : True
              is_inertia_regressor : True
              has_coupled_wrist : False
              embedded_forces : False
      processing_params:
          -   cut_off_frequency_butterworth: 100.0
              ts : 0.01
      tls_params:
          -   mass_load : None
              which_body_loaded : None  
              sync_joint_motion : False 
  

• Step 2: Generate sampled exciting postures and trajectories for experimentation.
  • For geomeotric calibration: Firstly, considering the infinite possibilities of combination of postures can be generated, a finite pool of feasible sampled postures in working space for the considered system needs to be provided thanks to simulator. Then, the pool can be input for a script optimal_config.py with a combinatorial optimization algorithm which will calculate and propose an optimal set of calibration postures chosen from the pool with much less number of postures while maximizing the excitation.
  • For dynamic identification: A nonlinear optimization problem needs to formulated and solved thanks to Ipopt solver in a script namde optimal_trajectory.py. Cost function can be chosen amongst different criteria such as condition number. Joint constraints, self-collision constraints should be obligatory, and other dedicated constraints can be included in constraint functions. Then, the Ipopt solver will iterate and find the best cubic spline that sastifies all constraints and optimize the defined cost function which aims to maximize the excitation for dynamics of the considered system.
• Step 3: Collect and prepare data in the correct format.
  To standardize the handling of data, we propose a sample format for collected data in csv format. These datasets should be stored in a data folder for such considered system.
• Step 4: Create a script implementing identification/calibration algorithms with templates. Dedicated template scripts calibration.py and identification.py are provided. Users needs to fill in essential parts to adapt to their systems. At the end, calibration/identification results will be displayed with visualization and statistical analysis. Then, it is up to users to justify the quality of calibration/identification based on their needs.
• Step 5: Update model with identified parameters.
  Once the results are accepted, users can update calibrated/identified parameters to their urdf model by scripts update_model.py or simply save to a xacro file for later usage.

• Calibration tools
• Identification tools
• Measurements
• Meshcat viewer
• Common tools

• cyipopt
• numdifftools
• quadprog
• numpy
• scipy
• pinocchio
• ndcurves
• hppfcl

• Pinocchio (source)
• Ipopt (source)