grasp-synergy repository

Repository Summary

Checkout URI
VCS Type git
VCS Version master
Last Updated 2019-12-08
CI status No Continuous Integration
Tags No category tags.
Contributing Help Wanted (0)
Good First Issues (0)
Pull Requests to Review (0)


Name Version
grasp_synergy 0.0.2


Grasp Synergy ROS package

Build Status

This package implements grasp synergies (aka eigengrasps): low-dimensional representation of hand configurations.

Very basically, grasp synergies are computed by doing PCA on a bunch of hand configurations (joint angles) to infer a low-dimensional representation.

Points in this low-dimensional space can then be reprojected to the full configuration space to control the hand.

This ROS package is independent of any hand configuration. It consists of two components: - the GraspSynergy class: compute & store synergies from high-dimensional grasp configurations, and compute grasps from low-dimensional input. - the synergy_node: ROS magic to provide nice interfaces for controlling grasp synergies.


Grasp Synergies (aka eigengrasps) are just low-dimensional representations of the grasp learned using PCA.

The GraspSynergy class is just a convenience class for computing, storing, and commanding grasp synergies. It is hand-agnostic.

Grasping data can be of any form: - A numpy matrix of joint values (N by D) - A list of sensor_msgs/JointState messages - A filepath to a rosbag file.

Here N are the number of collected grasps and D is the full dimensionality of the hand. We extract all D principal components (synergies), but in practice we will only use a smaller number.

To compute a grasp (in the original space), simply pass in a B-dimensional vector of coefficients to compute_grasp. (Nominally, B < D.) The method will automatically figure out how many components to use, and return the corresponding grasp configuration in the original D-dimensional space.


The synergy node enables you to load a grasp synergy space, create subscribers for the synergy space, and publish desired joint state messages to control the hand. It creates one top-level subscriber for fully-specified coefficient vector and many per-component subscribers for each component (each for a single value).

Each time the node receives a new low-dimensional (synergy-space) point, it computes the hand configuration and publishes a message with the desired joint states.

Subscribers: The node can create a variable number of subscribers: one for the fully-specified coefficient vector, and one per component (with singleton values). By default, the node is trained using a bag file of hand configuration data.

So for example, you might have the following subscribers for 5 synergies:

 * /grasp_synergy [std_msgs/Float32MultiArray]
 * /grasp_synergy/syn_0 [std_msgs/Float32]
 * /grasp_synergy/syn_1 [std_msgs/Float32]
 * /grasp_synergy/syn_2 [std_msgs/Float32]
 * /grasp_synergy/syn_3 [std_msgs/Float32]

This enables you to use tools like rqt_ez_publisher ( to provide a nice slider GUI interface to the synergy space:


NOTE: Due to a small bug in rqt_ez_publisher, if you want to use the sliders to control the individual synergies (/grasp_synergy/syn_N) you need to use the latest rqt_ez_publisher (checkout directly from github into your workspace and run the local copy).

Publisher: The desired joint state topic must be given.


No found.