Learning Controller Behaviour within the Control Basis Framework

Acronym: 
LECOBE
Research Areas: 
A
D
Abstract: 

LECOBE explores novel applications of the Control Basis Framework (CBF) idea to the problem domain of manual interaction. One milestone of the project is to develop an open source implementation of the CBF upon which further work is based. Furthermore the project investigates how to apply the CBF to the synthesis of grasps and the incorporation of tactile information into manual interactions in general.

Methods and Research Questions: 

Biological systems demonstrate a range of manual interactions that are not rivaled by any existing mechanical system. Only recently have anthropomorphic hands equipped with haptic sensors become available to the robotics research community. Creating controllers that allow the same range of interactions like e.g. humans is an interesting and challenging open research question.

One point of interest of the project is the integration of dimensionality reduction methods into the general framework to allow the generation of feasible and naturally looking grasps. Another is the novel technique of "tactile servoing" which is inspired by the well known visual servoing idea to the domain of tactile images. Tactile servoing is well suited to guide exploration and grasp synthesis.

Some of the open questions that we hope to be able to answer with this project are: What is a good dimensionality reduction method to constrain the motion of a robotic hand to the space of feasible and naturally looking interactions? And how to build a controller utilizing that representation? What are good tactile image features for tactile servoing? How to fuse the sensor information from pose estimation and haptic feedback to allow control?

An empirical study will provide the necessary pose data for the dimensionality reduction. Several algorithms will be compared and the feasibility of the produced representation will be investigated.

Outcomes: 

An open source implementation of the CBF has been developed. It has been successfully used in several applications inside of CITEC and CoR-Lab. Grasp synthesis and tactile servoing will be further investigated.

 

Publications: