- 1 Research
- 1.1 Nonlinear State Estimation
- 1.2 Multimodal Extended Range Telepresence
- 1.3 Intention Recognition
- 1.4 Information Processing in Sensor-Actuator-Networks
- 1.5 Stochastic Model Predictive Control
- 1.6 Miniature Walking Robots
- 1.7 Motion Compensation of Beating Heart
- 1.8 Tracking of an Operator's Movements
- 1.9 Highly Accurate Measurement of Calibration Parameters of Multi-Axes Machine Tools
Nonlinear State Estimation
Nonlinear state estimation is the core research field at ISAS. It provides methods for determining estimates from noisy measurements of only indirectly accessible states. Typical examples are localization tasks or reconstruction of distributed phenomena. Other applications are described in the sections below. No satisfying generic approaches exist for nonlinear systems with stochastic disturbances today, as no closed form expressions with bounded complexity for recursive processing tasks are known. Since recursive processing is needed for most technical estimation problems, approximations are indispensable for real world applications.
The ISAS is developing new estimators belonging to the class of analytic filters. They allow closed form expressions with constant or adjustable complexity. The key idea is to employ different classes of probability densities which are able to approximate arbitrary probability densities with arbitrary precision. Such classes include Dirac-mixtures, Gaussian-Mixtures, hybrid and orthogonal densities. For these classes, closed form estimators can be derived. Due to the use of distance measures for approximating given densities, the quality of the estimator can be adjusted with respect to computational complexity
A further approach investigated is the use of non-stochastic uncertainties in the form of sets of probability densities. These methods allow the systematic incorporation of modelling and numeric uncertainties as well as inevitable approximations of estimators.
More information on this topic can be found in our Publications.
Multimodal Extended Range Telepresence
We started our research in the field fo virtual reality (VR) as early as 2002. Our main focus lies on getting the best possible immersion and giving the user the ability to move around the virtual environment as freely as possible. The good immersion is achieved via realistic visual, acoustic and haptic feedback from the virtual world. With the addition of Motion Compression the user is able to move around much larger virtual spaces than the real working space would allow in a 1-to-1 maaping. Additionaly we are also doing research in the field of Augmented Reality (AR) since 2015.
current resarch topics VR:
- porting the Motion Compression algorithms to a HTC-Vive environment
- Motion Compression with better intention recognition
- Motion Compression with multiple users
- upgrading the haptic interface
current resarch topics AR:
- worker assitance systems in an industry 4.0 environment in cooperation with PFW Aerospace GmbH
- AR-glasses in orthopedic surgery
- calibration with high accuracy
- ergonomic presentation
completed research projects:
- Motion Compression (, , )
- Pedestrian Simulation (, , )
- Haptic Guidance (, , , )
- e-Installation ()
Contact: Christian Tesch
Human-robot cooperation is governed by mutual estimation of intentions. Providing intention recognition capabilities to technical systems will enable these to communicate implicitly with their human users. The human-robot cooperation research project of the SFB 588 "Humanoid Robots - Learning and Cooperating Multimodal Robots" is guided by this insight.
Intention recognition at the ISAS is based on a stochastic modelling approach using hybrid dynamic bayes nets. Bayes nets are cascaded stochastic models and represent causal relations between intentions, actions and observations explicitely. The focus of our research is on a system theoretic treatment of non-linear variable dependancies and hybrid scenarios, which contain continous as well as discrete random variables. More...
Information Processing in Sensor-Actuator-Networks
Proceedings in miniaturization of microprocessors, sensors, and actuators allows to embed small low-cost sensor-actuator-nodes with wireless communication into the environment. Self-organizing networks consisting of a multitude of such nodes provide the opportunity for developing novel applications, e.g. realtime mapping of pollution concentrations in cities.
The model-based techniques for sensor-actuator-networks developed at ISAS allow to reconstruct and identify complex distributed phenomena (like the pollution concentration) using just a small number of measurements. By systematically treating the appearing uncertainties, the information gain of future measurements is predictable. Through this, optimal sensor scheduling with respect to high measurement accuracy and low needs in energy consumption and communication is possible.
A further aspect in reducing the communication and computation costs as well as in efficiently applying sensor-actuator-networks concerns with the decentralized execution, i.e., the distributed execution over all nodes of the developed algorithms. Here, the consideration of stochastic dependencies is very challenging. Robust methods that explicitly model these dependencies are currently investigated at ISAS.
Stochastic Model Predictive Control
In Model Predictive Control (MPC), not only the current system state of a technical system, e.g. a mobile robot, but also the future system behavior, which depends on the possible control inputs, is considered in the control. This leads to a significant improvement in the quality of control. The Stochastic Nonlinear Model Predictive Control (SNMPC) methods developed at ISAS are especially well-suited for nonlinear systems that are heavily noise-influenced as they explicitly consider these aspects in the control, which is accomplished by integrating the nonlinear state estimation techniques developed at ISAS. The results are applied to Networked Control Systems and multi-agent systems.
The practical evaluation of the ISAS SNMPC methods is done with a team of miniature walking robots. Here, the focus is especially on the motion control, the path planning and the resource scheduling.
Miniature Walking Robots
At ISAS, a team of miniature walking robots is been developed. The robots have six independent degrees of freedom, which allow them to move in a wide variety of motion patterns. These motion patterns comprise rotation, sideward and forward movement as well as a combination of all three of them. The robots are radio controlled, have a high-performance rechargeable battery, and integrated computing resources for motion control. For various experiments from the fields of nonlinear state estimation (e.g. cooperative position estimation), resource scheduling (e.g. measurement scheduling) as well as model predictive control (e.g. path planning), the robots are integrated into a test-environment. In this test-environment, the robots’ poses are tracked by an overhead camera, which allows to simulate a wide variety of different sensors like, e.g. ultrasonic transducers for distance measurement
Motion Compensation of Beating Heart
| In order to assist surgeons at minimal invasive interventions on the beating heart it would be helpful to develop a robotic surgery system, which synchronizes the instruments with the beating heart, so that their positions do not change relatively to the point of interest (POI). When the surgical robot takes care of the synchronization, the provided visual stabilization enables the presentation of the beating heart as if it was stationary and non-moving.
For synchronization of the surgical instruments with the POI and for the visual stabilization, the heart wall is regarded as a viscoelastic physical body. Its motion is described by a distributed parameter system mathematically formulated by a system of stochastic partial differential equations. By employing an element-free method for spatial discretization and a time integration method for temporal discretization, this system is converted in a state-space form. The internal parameters and states of the system are estimated by means of stochastic estimation approach, which processes discrete-time and discrete-space noisy camera measurements. As a result, the spatially and temporally distributed heart wall motion is reconstructed for arbitrary POIs even when no measurement information is available.
The project is supported by the Research Training Group GRK 1126 "Intelligent Surgery".
Tracking of an Operator's Movements
For capturing the movement of persons, which is e.g. necessary in Large-Scale Telepresence, an acoustic tracking system is being developed at the ISAS. Multiple stationary loudspeakers simultaneously broadcast different signals, which are then captured by microphones attached to the persons and assigned to their respective sources. The posture of the person is estimated based on delay, loudspeaker positions and microphone positions. By using broadband signals, the system is robust against shadowing effects (i.e. people or furniture). A high level of accuracy is attained by using of a large number of microphones and speakers as well as modern signal processing and localization methods. With a newly developed localization method the acquisition rate at the same sampling frequency will be increased while retaining the same degree of precision as before. In this way, fast body movements can also be tracked reliably.
Contact: Frederik Beutler
Highly Accurate Measurement of Calibration Parameters of Multi-Axes Machine Tools
In cooperation with Prueftechnik Alignment Systems GmbH a measuring system to calibrate multi-axes machine tools is beeing developed. Objective is to replace conventional methods, which use dial gauges and gauge steel, by a laser based method. The new system will reduce the time needed for a machine calibration from a couple of days to a few hours. On one hand, an instrument is being designed, which will measure deviations within micrometers respectively microdegrees. One the other hand research on algorithms is done, which on the bases of a machine model determine out of a minimum of measuring points the calibration parameters. Considering model and measuring uncertainties as well as the prediction of optimal measuring positions will ensure economic usability. These algorithms will be the base for general concepts of sensor deployment planing.
Contact: Dietrich Brunn