Only released in EOL distros:
Package Summary
This package contains the core library for kinematic model fitting for articulated objects. Examples of articulated objects are gangway or cabinet doors, door handles, and drawers. It provides both a C++ library for model fitting and selection, and nodes that make the functions of this library available as ROS services to other nodes.
- Author: Juergen Sturm
- License: BSD
- Repository: alufr-ros-pkg
- Source: svn https://alufr-ros-pkg.googlecode.com/svn/tags/stacks/articulation/articulation-0.1.3
Package Summary
This package contains the core library for kinematic model fitting for articulated objects. Examples of articulated objects are gangway or cabinet doors, door handles, and drawers. It provides both a C++ library for model fitting and selection, and nodes that make the functions of this library available as ROS services to other nodes.
- Author: Juergen Sturm
- License: BSD
- Source: svn http://alufr-ros-pkg.googlecode.com/svn/trunk/articulation
Contents
Tutorials on Articulated Objects
- Learning Kinematic Models for Articulated Objects using a Webcam
This tutorial is a step-by-step guide that instructs users how to learn kinematic models of articulated objects only by using a webcam and a laptop.
- Getting started with Articulation Models
This tutorial guides you step-by-step through the available tools for fitting, selecting and displaying kinematic trajectories of articulated objects. We will start with preparing a text file containing a kinematic trajectory. We will use an existing script to publish this trajectory, and use the existing model fitting and selection node to estimate a suitable model. We will then visualize this model in RVIZ.
- Using the Articulation Models (Python)
In this tutorial, you will create a simple python script that calls ROS services for model fitting and selection. The script will output the estimated model class (like rotational, prismatic, etc.) and the estimated model parameters (like radius of rotation, etc.).
- Using the Articulation Model Library (C++)
This tutorial demonstrates how to use the articulation model library directly in your programs. This is more efficient than sending ROS messages or ROS services. In this tutorial, a short program is presented that creates an artificial trajectory of an object rotating around a hinge, and then uses the model fitting library to recover the rotational center and radius. Further, the sampled trajectory and the fitted model are publishes as a ROS message for visualization in RVIZ.
- Learning Kinematic Models from End-Effector Trajectories
This tutorial demonstrates the process of model fitting and model selection to real data recorded by a mobile manipulation robot operating various doors and drawers.
Model Classes and Model Parametrization
At the moment, the following model classes are implemented:
rigid: a rigid model, describes a static link (with Gaussian noise)
prismatic: a prismatic joint model, describes for example a drawer or a sliding cabinet door
rotational: a rotary joint model, describes for example a door or a door handle
pca_gp: a non-parametric model, capable of describing many other articulation models, like garage doors, or combinations of sliding and rotary motion.
All nodes in this package query the ROS parameter filter_models, which names the model classes that will be considered during model selection. The default is rigid prismatic rotational. Globally, all model classes accept the parameters sigma_position and sigma_orientation, that define the properties of the Gaussian noise process.
Model Parameters
The model estimators fit the models to the data, and store the found parameters in the ModelMsg::params vector.
The parameters of the rigid model are:
rigid_position[3], gives the estimated position of the rigid object.
rigid_orientation[4], gives the estimated orientation of the rigid object.
rigid_width, gives the estimated width of the rigid object (only if a width channel is present in ModelMsg::track.channels).
rigid_height, gives the estimated height of the rigid object (only if a height channel is present in ModelMsg::track.channels).
The parameters of the prismatic model are:
rigid_position[3], gives the average position of the articulated object (inherited from the rigid model)
rigid_orientation[4], gives the average orientation of the articulated object (inherited from the rigid model)
prismatic_dir[3], gives the line direction of the prismatic joint
The parameters of the rotational model are:
rot_center[3], points to the rotational center of the rotational joint
rot_axis[4], gives the rotational axis of the rotational joint
rot_radius, gives the radius of the rotational joint
rot_orientation[4], gives the orientation of the articulated object
The pca_gp model is parameter-free, but requires the data points of the trajectory to make predictions.
Nodes
model_learner_msg
This node receives kinematic trajectories in form of articulation_msgs/TrackMsg messages, fits the trajectories to all known model classes, selects the model with the lowest BIC (Bayesian Information Criterion), and publishes the result in form of a articulation_msgs/ModelMsg message. The node operates in real-time i.e., it has a queue size of 1 and might drop messages if processing takes too long.Subscribed Topics
track (articulation_msgs/TrackMsg)- a kinematic trajectory of the articulated object, as observed by the robot
Published Topics
model (articulation_msgs/ModelMsg)- the resulting articulation model, after model fitting and selection
Parameters
~sigma_position (float, default: 0.01)- This parameter specifies the standard deviation of the assumed Gaussian noise in the Cartesian position of the observed trajectories. Unit: meters.
- This parameter specifies the standard deviation of the assumed Gaussian noise in the orientation of the observed trajectories. Unit: radians.
- This parameter specifies which model classes are considered during model fitting and model selection. See the section on model classes for a complete list of available models. See the tutorial for how to add custom model classes.
model_learner_srv
This node provides ROS services for model fitting, selection and evaluation. This nodes provides high-level access to most of the functionality of articulation_models C++ library.Services
model_fit (articulation_msgs/TrackModelSrv)- (Re-)fits the parameters of the model to the given trajectory. The model class needs to be set in the request.model.name field. The node then evaluates the model given these parameters and returns the resulting model. The model parameters, and the result of the evaluation, is stored in the response.model.params list. The projection of the trajectory, given the fitted model parameters, is stored in response.model.track.pose_projected.
- Selects the model class, given a noisy, unclassified trajectory of an articulated object. The node fits the given trajectory to all available model classes, evaluates the candidate models and selects the model with the lowest BIC (Bayesian Information Criterion). The resulting model is evaluated, and returned in response.model.
- Evaluates a trajectory, given a model class and the model's parameters. In particular, the node computes the data likelihood, the BIC value, etc. Further, it projects the trajectory onto the model (response.model.track.pose_projected),
Parameters
~sigma_position (float, default: 0.01)- This parameter specifies the standard deviation of the assumed Gaussian noise in the Cartesian position of the observed trajectories. Unit: meters.
- This parameter specifies the standard deviation of the assumed Gaussian noise in the orientation of the observed trajectories. Unit: radians.
- This parameter specifies which model classes are considered during model fitting and model selection. See the section on model classes for a complete list of available models. See the tutorial for how to add custom model classes.
model_learner_prior
This node is an extension of the model_learner_srv node. Next to providing ROS services for model fitting, selection and evaluation, and allows additionally to store, retrieve prior models and use them during model fitting and selection. Furthermore, it provides a service call for evaluating the model fitting/selection of partial trajectories, without the need for repeatedly calling the model selection service.Services
model_select (articulation_msgs/TrackModelSrv)- Selects the model class, given a noisy, unclassified trajectory of an articulated object. The node fits the given trajectory to all available model classes, evaluates the candidate models and selects the model with the lowest BIC (Bayesian Information Criterion). The resulting model is evaluated, and returned in response.model. If a model class name is provided in request.model.name, only the parameters of this model are (re-)fitted. The model parameters, and the result of the evaluation, is stored in the response.model.params list. The projection of the trajectory, given the fitted model parameters, is stored in response.model.track.pose_projected.
- Evaluates a trajectory, given a model class and the model's parameters. In particular, the node computes the data likelihood, the BIC value, etc. Further, it projects the trajectory onto the model (response.model.track.pose_projected).
- Store a model in the prior model database. The model is identified uniquely by its id (request.model.id), so repeated calls to model_store with the same model.id will overwrite the corresponding model. A special value of model.id = -1 will automatically assign an unused id value to the field. This automatically assigned identifier can then be retrieved in response.model.id.
- Retrieves the current list of prior models, stored in the model_learner_prior node. Useful for inspection, or saving the prior to disk (see prior_save.py for an example)
- Sets the current list of prior models in the model_learner_prior node. This service can also be used to clear the list of known prior models, simply by calling it with an empty list. Useful for restoring the prior from disk (see prior_load.py).
Parameters
~sigma_position (float, default: 0.01)- This parameter specifies the standard deviation of the assumed Gaussian noise in the Cartesian position of the observed trajectories. Unit: meters.
- This parameter specifies the standard deviation of the assumed Gaussian noise in the orientation of the observed trajectories. Unit: radians.
- This parameter specifies which model classes are considered during model fitting and model selection. See the section on model classes for a complete list of available models. See the tutorial for how to add custom model classes.
- When prior models are available, this parameter specifies whether the new data trajectory is aligned to the existing trajectory data of the prior model using the Iteratively Closest Point algorithm (see also <<TracLink(ros-pkg IPC)>>).
Useful Scripts
prior_publisher_atonce.py
This node reads trajectories from multiple files, calls model fitting and selection, and stores the found model as a prior. For an example, run roslaunch demo_fitting prior_iros_at_once.launch.Published Topics
model (articulation_msgs/TrackMsg)- Resulting models, after model fitting and selection. Useful for visualization in RVIZ.
Services Called
model_select (articulation_msgs/TrackModelSrv)- Select a suitable model for the trajectory. }
- Store the found model as a prior model, and use it as prior information during model fitting and selection for future trajectories.
simple_publisher.py
This node publishes a trajectory from a text file. Each line of the text file contains a single pose, with space-separated values. The format of a pose is "x y z [qw qx qy qz [width height]]", where x y z is the Cartesian position, qw qx qy qz is the (optional) orientation, and width and height is the (optional) dimensions of a door/drawer front. The script reads the complete file at once, then moves the trajectory to start at the origin. Then it incrementally sends out articulation_msgs/TrackMsg messages, at 0.05 second intervals. For an example of this script, run roslaunch demo_fitting fit_kitchen.launch.Published Topics
track (articulation_msgs/TrackMsg)- Incrementally publishes kinematic trajectories that it reads from a text file. }
prior_save.py
This node retrieves the current prior models using the model_prior_get service and stores it in a file.Services Called
model_prior_get (articulation_msgs/GetModelPriorSrv)- Retrieves the current list of prior models. }
prior_load.py
This node loads prior models from a file, and restores them using the model_prior_set service.Services Called
model_prior_set (articulation_msgs/SetModelPriorSrv)- Sets the current list of prior models. }
References
Jürgen Sturm, Cyrill Stachniss, Wolfram Burgard. A Probabilistic Framework for Learning Kinematic Models of Articulated Objects. In Journal on Artificial Intelligence Research (JAIR), volume 41, 2011. [ pdf ] [ bibtex ]
Jürgen Sturm, Advait Jain, Cyrill Stachniss, Charlie Kemp, Wolfram Burgard. Operating Articulated Objects Based on Experience. In Proc. of the International Conference on Intelligent Robot Systems (IROS), Anchorage, USA, 2010. [ pdf ] [ bibtex ]
Jürgen Sturm, Kurt Konolige, Cyrill Stachniss, Wolfram Burgard. Vision-based Detection for Learning Articulation Models of Cabinet Doors and Drawers in Household Environments. In Proc. of the IEEE International Conference on Robotics and Automation (ICRA), Anchorage, USA, 2010. [ pdf ] [ bibtex ]
Jürgen Sturm, Cyrill Stachniss, Vijay Pradeep, Christian Plagemann, Kurt Konolige, Wolfram Burgard. Learning Kinematic Models for Articulated Objects. In Proc. of the International Joint Conference on Artificial Intelligence (IJCAI), Pasadena, USA, 2009. [ pdf ] [ bibtex ]
More Information
More information (including videos, papers, presentations) can be found on the homepage of Juergen Sturm.
Report a Bug
If you run into any problems, please feel free to contact Juergen Sturm <juergen.sturm@in.tum.de>.
