|
Ongoing research projects
|
| CogMaSh - Cognitive Machine Shop | | # |
CogMaSh is a part of the TUM
Excellence Initiative cluster CoTeSys and aims at
elevating the until now applied techniques which make Flexible Manufacturing Systems (FMS) out of ordinary
Production Plants. The break-through concept that we want to apply comes behind the word Cognition, i.e. the
flexibility, adaptivity and generality in FMSes will be reached through human-like perception, reasoning, learning,
adapting and process planning. That particular FMS shall in aftermath autonomously cope with different context
scenarios and situations, assure sufficient safety issues and after all significantly reduce production costs.
In my work I am dealing with sensory embodied in suchlike FMSes which gives rise
to intelligent perception paradigms and provides grounds for better and
above all certain reasoning of going-on actions. Image on the left hand side
shows our two testbed setups, a real-hardware FMS as installed by provider, and
in the right part of the image, an in-house built-up FMS using the Gazebo
3D simulator (work of Thomas Ruehr). My work is currently concentrated around the latter one
because it allows for unrestricted access to FMS's kinematics and control
structures as oppose to the proprietary software driving the real FMS.
In
Gazebo, I am currently extracting the infrastructure by building-up such
models as the below shown RA01 robotic arm which is to be in the future
manufactured by our FMS all alone. In addition I am/will be also providing
Player drivers and belonging functionalities of the sensors in use. In the
middle-term run I aim at estimating the probabilistic state of the FMS,
reason based on the outcome combined with the past knowledge and undertake
the corrective actions if needed. The production processes run in the plan-based
manner (using Reactive Plan Language) and have the ability to get transformed
in order to improve their efficiency.
|
Relevant publications
- Thomas Ruehr, Dejan Pangercic, and Michael Beetz.
Structured Reactive Controllers and Transformational Planning for Manufacturing.
In Proceedings of The 13th IEEE International Conference on Emerging Technologies and Factory Automation (ETFA), Hamburg, Germany, September 15-18, 2008.
[PDF]
|
| Monocular 3D SLAM for Indoor Environments | | # |
Under the shared supervisory of Radu Bogdan Rusu,
Marko Munih and Michael Beetz
I wrote my Master Thesis on 3D-based monocular SLAM and applied it to the
Assistive Kitchen environment. The abstract right
under here as well as the PDF and the final movie of the work in thesis done are all available below.
Visual 3D SLAM (vSLAM) is inarguably the red-hot trend in probabilistic robotics.
By answering the two key robot-questions of where I am, and where I am going, it
solves the famous SLAM 'chicken-egg' problem. The paid-off of using vision as an
only outward-directed robot sense is in that it utilizes camera inherent properties to
return very intuitive and well-understood information. As such it outperforms any
kind of other sensors (laser, odometry, sonar). The work done in this thesis tackles
with vSLAM using only one camera which is attached to an omnidirectional robot and
delivers eight bit, monochrome images. The robot operates in an assistive kitchen environment,
which is artificially cluttered to enrich its texture. A sparse map in 3D
space of the rich-texture features of size 11 × 11 pixels is being built by using an Extended
Kalman Filter state-estimator, while robot moves and concurrently uses this
map for the navigation. The important challenge, solved in this work, is a feature
3D depth estimation using a 1D Particle Depth Estimation algorithm, which after
all enables the utilization of the monocular camera. The algorithm currently SLAMs
table-size volumes and runs in realtime on the standard laptop machine. The underlying
essence of this work is in its mathematical inductions that perfectly discuss robot
motion model and measurement model transformations and respective covariances.
Applied measurement model is a novel approach and has been completely developed by
us.
VIDEO:
|
Relevant publications
- Pangercic Dejan, Radu Bogdan Rusu, and Michael Beetz.
3D-Based Monocular SLAM for Mobile Agents Navigating in Indoor Environments.
In Proceedings of The 13th IEEE International Conference on Emerging Technologies and Factory Automation (ETFA), Hamburg, Germany, September 15-18, 2008.
[PDF]
- Pangercic Dejan, Radu Bogdan Rusu,
Marko Munih, and Michael Beetz.
Monocular 3D SLAM for Indoor Environments. Master thesis, TU Muenchen, December 2007.
[PDF]
|
|
|
Past/finished researched projects
|
| 7IR - Seven Days of Industrial Robotics | | # |
As a member of the EESTEC association I, when I was still an undergraduate
student, together with ROBO Laboratory, companies
Motoman and ABB organized a break-through
week-event on industrial robotics intended for the freshman and sophomore
students at the FE Ljubljana.
Practical tutorials on a)joint robot-robot actions, b)pick and place
and screw robot-vision tasks, c)planning robot's trajectories and understanding
its forward and inverse kinematics, d)off-line simulator programming and all
that topped with the e)robot soccer tactical assignment were a definite GO for
Robotics studies in the number of participating students. A collection of
tutorial instructions prefaced with an excellent introduction to the industrial
robotics world is available below.
|
|
| My Undergrad Projects | | # |
For the list of Projects (and more) I did during my studies at the TUM and
FE please visit my personal website.
|
|
|
|
|
|
|