Thesis

General Information:

Students are welcome to produce their Bachelor or Master thesis in the CoR-Lab. You find below suggested topics. Students can propose their own topics of interest, which will be formated to a proper and tractable set of research questions in an initial discussion.

Bachelor Theses:

Comparison of a Neural Reaching Movement Controller to Human Reaching Data

This thesis will compare movements generated by a neural network controller to human reaching movements. In a first step, velocity profiles and their properties like symmetry and shape have to be compared to properties of human reaching movements by means of statistical methods. This includes the falsification whether a velocity profile is bell-shaped or not. In a second step, the neural controller can be extended to explicitly parameterize the velocity profile shape. Matlab or C++ sources of the neural controller are provided to the applicant.

Collision Detection for the Humanoid Robot iCub

The main goal of this thesis is the implementation of a collision detection module for the iCub. This requires the combination of existing code for the geometric representation of the robot kinematics with available collision detection algorithms from the computer graphics community. The focus of this thesis resides on the software engineering.

Selecting the best home-posture: flexible redundancy resolution with Goal Babbling

Redundant sensorimotor systems allow to perform reaching movements in a variety of ways. During autonomous sensorimotor exploration with "Goal Babbling", such resolution of redundancy can be determined, and exploited, by means of a home-posture: a resting-point from which the learner develops its coordination repertoire. However, it is not always obvious how good home-postures should look like. The main goal of this thesis is to find criteria to select a good home-posture for a given sensorimotor coordination problem. In a second step these criteria can be evaluated on problems that benefit from switching between different redundancy resolutions, such as obstacle avoidance, or spaces that can not be entirely reached from a single home posture. C++ source code will be provided to the applicant.

Implicit obstacle detection from autonomous sensorimotor exploration

If obstacles prevent certain reaching movements, the standard approach is to detect them by visual perception, or avoid them on haptic contact. But what is the impact of obstacles on the actual sensorimotor coordination, and in particular on on sensorimotor learning? The goal of this thesis if to explore the impact of obstacles on autonomous sensorimotor exploration with "Goal Babbling". The central idea is that obstacles might prevent certain movements towards a target, while other movements might still be possible. The hypothesis to be investigated by the applicant is: do the exploratory movements generated by Goal Babbling allow to infer shape or position of an obstacle? C++ source code will be provided to the applicant.

Master Theses: