Research Topics for Students
If students are looking for an interesting BSc, MSc or PhD research project along any of the broad research lines below, they are encouraged to get in touch.
 |
|
Robot control for recharging
Intelligent robotics is a wide area covering vision, control and bio-inspired learning. In this thesis, a robot action sequence is developed using reinforcement learning. The humanoid robot Nao shall recharge its batteries without the need of an operator.
Goals:
Requirements:
[1] Goal-Directed Feature Learning. Weber & Triesch (2009),
www.informatik.uni-hamburg.de/~weber/publications/09IJCNN_RLfeatures_0660.pdf
For further questions contact any of the below.
Contact:
Professor Dr. Stefan Wermter, Dr. Cornelius Weber, Nicolas Navarro
|
|
|
Person recognition from a ceiling camera
The EU project KSERA aims at helping elderly persons at home. For this purpose, a
ceiling camera localises a helping robot and a person. To detect a person from a
ceiling camera with a fisheye lense is challenging, because the person's appearance
is very different seen from the top or from a side. We wish to implement a semisupervised
classification algorithm for this task [e.g. 1].
Goals:
Requirements:
[1] Goal-Directed Feature Learning. Weber & Triesch (2009),
www.informatik.uni-hamburg.de/~weber/publications/09IJCNN_RLfeatures_0660.pdf
For further questions, please contact any of the below.
Contact:
Professor Dr. Stefan Wermter, Dr. Cornelius Weber, Dipl.-Ing. Alex Yan|
|
|
Robot Grasping
Intelligent robotics is a wide area ranging from bio-inspired learning, vision, motor control to household assistance. A central task that has been performed with robots but that still needs more development is to grasp an object. This thesis will tackle that task using the humanoid-shaped robot Nao and using learning neural networks. There are two sub-tasks:
(i) to align the robot close to the object by walking and
(ii) to grasp the object via arm movement and hand closure.
Each of these sub-tasks alone, successfully implemented, warrants one whole thesis
Goals:
Useful Skills:
Contact:
Dr. Cornelius Weber, Professor Dr. Stefan Wermter, Nicolas Navarro
|
|
|
Bio-inspired auditory pathway
Robots need to be aware of their environment even if important things happen outside of their visual input field. A key factor is to move
from simulation environments to embodied applications and architectures that present different kinds of challenges such as environmental noise
and even ego-noise. In this project, we will develop a bio-inspired auditory model for our new iCub head.
This system will mimic bottom-up auditory signal processing integrating known biological models into a single framework.
The model will match time and degree of processing constraints and will act as the core architecture for developing future
bio-constrained and bio-inspired models of sensory signal processing in the brain.
Please note that you don’t need to address all the following listed features and they can be chosen depending on your interest:
The model should include features such as a cochlear model, auditory adaptation (dynamic background noise filtering), ego-noise filtering,
signal onset and end detection, auditory input classification or segmentation, sound source localization, etc.
General goals:
Useful skills:
Contact:
Professor Dr. Stefan Wermter, Dr. Cornelius Weber, Nicolas Navarro
Email: {wermter,weber,navarro}@informatik.uni-hamburg.de
Telefon: +49 40 42883-2434, -2537, -2530
|
|
|
Bio-inspired image stabilizing on a walking robot
Intelligent robotics ranges from bio-inspired learning, vision, motor control to
household assistance. A key factor is to move from simulation environments to
embodied applications, which presents different kinds of challenges such as noisy
sensory signals and system perturbations. For example, the shaking video taken
from a walking robot’s camera is unpleasant to watch, however, when we walk, we
perceive a stable environment. In this project, we will develop a bio-inspired digital
image stabilization system to be used on small-sized humanoid robots such as NAO
and Darwin-OP. The thesis can first address work on a video and then on a robot in
real-time.
Goals:
Useful skills:
Contact:
Professor Dr. Stefan Wermter, Dr. Cornelius Weber, Nicolas Navarro
|
|
|
Exploring the generalisation capabilities of a multiple timescale recurrent neural network (MTRNN)
Continuous time recurrent neural networks (CTRNNs) are a method for learning and generalizing sequences, for example a motor movement of a robot arm or a spoken utterance in natural language. In this project we want to explore the generalisation capabilities of a novel CTRNN which keeps some information from the past (some recurrent information) with different levels of timescale, namely an MTRNN. The different timescales of parts of the network can lead to a hierarchy of features of a sequence. Since the network has a rich number of parameters, many effects have not been explored yet and provide a very interesting bachelor project.
For a thesis on a masters level the project can be extended by a deeper analysis of the explored MTRNNs. For example with a principle component analysis (PCA) characteristics, i.e. the similarities of parts of the network for specific sequences can be detected and interpreted.
Goals:
Useful skills:
For further questions, contact Stefan Heinrich.
Contact:
Dipl.-Inform. Stefan Heinrich,
Professor Dr. Stefan Wermter
|
|
|
Speech Recognition with Long-Short-Term Memory Neural Networks
Automated speech recognition (ASR) is one of the key applications of computer algorithms and intelligent systems. Still, the state of the art of ASRs is based on Hidden Markov Models (HMMs) with more than 20 years of development and research. However, these systems are fairly specialised and already reached the pinnacle of success. Alternative methods exist that are computationally more powerful and biologically more plausible than HMMs, because they include continuous internal states. An example is the Long-Short-Term Memory (LSTM) neural network. For a bachelor thesis we want to build up an LSTM-based ASR following the experiments by Graves et al. 2004 and evaluate the recognition rate [1]. For a master thesis we also want to install the system on a humanoid robot NAO and compare the recognition results with a state-of-the-art system, e.g. the Sphinx ASR.
Goals:
Useful skills:
For further questions, contact Stefan Heinrich.
[1] Graves, A., Eck, D., Beringer, N. & Schmidhuber, J. Biologically plausible speech recognition with LSTM neural nets. Biologically Inspired Approaches to Advanced Information Technology 21, 127-136 (2004).
Contact:
Dipl.-Inform. Stefan Heinrich,
Professor Dr. Stefan Wermter
|
|
|
Neural network learning with a laser robot head
Being able to cover an indoor environment up to 4 meters, the
laser head of NAO provides a compact and accurate method to
obtain a 2D obstacle map for the use of humanoid robot
navigation. Moreover, the laser data is integrated in NaoQi and
can be easily retrieved by simple commands. This kind of laser
head has been initially developed as a stair climbing robot, an
obstacle avoidance robot, etc. In this project, we propose to
use the laser sensor for moving object prediction in an
ambient environment. While there is previous research [1] of
dynamic collision avoidance or moving object tracking, we
propose to realize these two aspects in a humanoid robot with
neural networks technologies.
Goals:
Useful skills:
Contact:
Professor Dr. Stefan Wermter, Dr. Cornelius Weber, Junpei Zhong
|
|
|
Further indicative projects in computing:
Creating a neuroscience-inspired learning interacting robot
In this exciting project we look at neuroscience-inspired architectures
for controlling the behavior of a robot. While traditional
robots have often been preprogrammed, our new approach will focus on learning
robots which will be based on some neuroscience evidence. These navigation
and movement concepts are transferred and further developed
on a Nao robot which has some speech and vision
capabilities. The NeuroBot will
learn to associate actions with words and pointing gestures. We want to restrict
the fixed manual programming of the robot and emphasize the adaptive autonomous
learning in neural networks in combination with restricted instructions
via words and simple pointing. Some of the previous and current robot platforms available are
shown at http://www.informatik.uni-hamburg.de/WTM/neurobots.
Students who participated in our project "Human Robot Interaction" can also
suggest a further topic for their theses.
Goals:
- Learning navigation from simple multimodal input
- Robotic vision: Object recognition and object manipulation
- Implementation of neural network algorithms for speech and pointing instructions for NeuroBot
- At later stage: vision and speech capabilities for the NAO
Requirements:
- Programming skills: C/C++, Python
- At least basic knowledge in neural network algorithms and natural language processing
- Willingness to work in robotic environment
The thesis can be written either in German or English. All topics could be tailored to be at a bachelor, master level, or phd student level. If you are interested contact us for discussion:
Contact:
Professor Dr. Stefan Wermter, Dr. Cornelius Weber, Dipl.-Inform. Stefan Heinrich|
|
|
Internet text mining agents based on hybrid and neural learning techniques
Unrestricted potentially faulty text messages arrive at a certain
delivery point (e.g. email address or world wide web address). These text
messages are scanned and then distributed to one of several expert agents
according to a certain task criterion (e.g. language class, news group
or semantic library class). Expert agents can dynamically accept
or reject a message. Furthermore, the number of expert agents
can vary dynamically over time. If the best possible expert agent
is not available the next best choice has to be found in order to supply
a best possible instant answer. This dynamic unrestricted message
routing task is new and hybrid neural/symbolic techniques have not
yet been examined for noisy dynamic internet message routing.
Goals:
- Development of hybrid neural and symbolic processing methods
- Implementation of an hybrid learning agent for text mining
Requirements:
- Programming skills: C/C++, Python, Matlab
- At least basic knowledge in neural networks/symbolic processing and mining techniques
- Willingness to explore data with algorithmic methods
The thesis can be written either in German or English. All topics could be tailored to be at a bachelor, master level, or phd student level. If you are interested contact us so for discussion:
Contact:
Dipl.-Inf. Doreen Jirak, Professor Dr. Stefan Wermter|
|
|
Neural and fuzzy state machines for intelligent understanding
Hybrid systems are powerful systems from
the point of view of knowledge engineering and cognitive science. Neural
processing supports for instance learning, robustness and graded plausible
interpretation, while symbolic fuzzy processing has advantages e.g. for
easy understanding, simple representation and efficient encoding in smaller
domains. One particularly new approach is based on the
theory of neural preference classes which can be interpreted as neural
network preferences or fuzzy symbolic preferences.
Goals:
- Coupling of neural network and fuzzy techniques
- Development of neural and fuzzy methods for algorithmic learning
- Implementation of neural/fuzzy algorithms based on neural preference classes
- In particular: Building a natural semantic system based on recurrent networks and symbolic fuzzy transducer
Requirements:
- Programming skills: C/C++, Python, Matlab
- At least basic knowledge in hybrid neural systems and mining techniques
- Willingness to explore data with algorithmic methods
The thesis can be written either in German or English. All topics could be tailored to be at a bachelor, master level, or phd student level. If you are interested contact us for discussion:
Contact:
Dipl.-Inf. Doreen Jirak, Professor Dr. Stefan Wermter|
|
|
Building integrated miners using knowledge extraction and neural networks
Documents like medical records, newspapers or
documents in the World Wide Web often contain structured data as well as
unstructured free text. In the future it will be even more essential to
use learning and data mining techniques, for instance from neural networks
and statistics, to acquire, process, and adapt knowledge from structured
and unstructured data.
Goals:
- Design and building of prototype system for data- and text mining and information extraction
- Combining several hybrid neural architectures and methods
- Development and Implementation of hybrid learning toolbox for mining purposes
- Evaluation of developed methods on the same datasets and text material
Requirements:
- Programming skills: C/C++, Python, Matlab
- Basic knowledge in hybrid neural systems and mining techniques
- Understanding of algorithmic methods regarding neural networks
- Motivation to work in the area of data analysis and data evaluation with statistics
The thesis can be written either in German or English. All topics could be tailored to be at a bachelor, master level, or phd student level. If you are interested contact us for discussion:
Contact:
Dipl.-Inf. Doreen Jirak, Professor Dr. Stefan Wermter|
|
|
Reccurent Neural Networks for Speech Recognition on Humanoid Robots (BSc Thesis) |
||
| Grounding of Language in Sensorimotor World Interaction of a Humanoid Robot, using Neural Networks (BSc Thesis) |
||
| Evolving Neural Networks for Bipedal Walking of a NAO Robot (BSc Thesis) |
||
| Geräuschquellenlokalisation mit einem humanoiden Roboter (BSc Thesis) |
