Contributions published at Artificial Intelligence Lab

In the field of developmental robotics, a lot of attention has been devoted to algorithms that allow agents to build up skills through sensorimotor interaction. Such interaction is largely affected by the agent's morphology, that is, its shape, limb articulation, as well as the position and density of sensors on its body surface. Despite its importance, the impact of morphology on behavior has not been systematically addressed. In this paper, we take inspiration from the human vision system, and demonstrate using a binocular active vision platform why sensor morphology in combination with other properties of the body, are essential conditions to achieve coordinated visual behavior (here, vergence). Specifically, to evaluate the effect of sensor morphology on behavior, we present an information-theoretic analysis quantifying the statistical regularities induced through sensorimotor interaction. Our results show that only for an adequate sensor morphology, vergence increases the amount of information structure in the sensorimotor loop.

Complex systems involving many interacting components being out of equilibrium often organize into patterns. Understanding the underlying principles that govern such systems might lead to a deeper insight into living systems and the development of new applications in robotics. In this contribution, we investigate water-based self-assembling modules, exhibiting a segregation effect under some particular conditions. The system consists of vibrating (active) and non vibrating (passive) circular modules floating on the surface of the water. The segregation happens as a result of a depletionlike force, which is of purely entropic nature and is based on the characteristics of the modules (active or passive). We focus especially on the dynamics of the process with respect to the energy and the entropy. Some applications of the designed system are also discussed.

The lonely researcher trying to crack a problem in her office still plays an important role in fundamental research. However, a vast exchange, often with participants from different fields is taking place in modern research activities and projects. In the "Research Value Chain" (a simplified depiction of the Scientific Method as a process used for the analyses in this paper), interactions between researchers and other individuals (intentional or not) within or outside their respective institutions can be regarded as occurrences of Collective Intelligence. "Crowdsourcing" (Howe 2006) is a special case of such Collective Intelligence. It leverages the wisdom of crowds (Surowiecki 2004) and is already changing the way groups of people produce knowledge, generate ideas and make them actionable. A very famous example of a Crowdsourcing outcome is the distributed encyclopedia "Wikipedia". Published research agendas are asking how techniques addressing "the crowd" can be applied to non-profit environments, namely universities, and fundamental research in general. This paper discusses how the non-profit "Research Value Chain" can potentially benefit from Crowdsourcing. Further, a research agenda is proposed that investigates a) the applicability of Crowdsourcing to fundamental science and b) the impact of distributed agent principles from Artificial Intelligence research on the robustness of Crowdsourcing. Insights and methods from different research fields will be combined, such as complex networks, spatially embedded interacting agents or swarms and dynamic networks. Although the ideas in this paper essentially outline a research agenda, preliminary data from two pilot studies show that nonscientists can support scientific projects with high quality contributions. Intrinsic motivators (such as "fun") are present, which suggests individuals are not (only) contributing to such projects with a view to large monetary rewards.

Developmental robotics focuses on how to endow robots with adaptive capabilities. Even though embodiment has been recognized as an essential factor for understanding development, there is yet not much work that investigates how the morphology of sensors and actuators shapes adaptivity and learning processes. Moreover, these studies are largely at an intuitive and qualitative level. In this paper, we address the issue by studying how in an active vision system sensor morphology and bodily features affect a behavior such as vergence. Specifically, we present an information-theoretic analysis of two experiments showing how adequate sensor morphology influences statistical dependencies in the sensorimotor loop. The results show that an appropriate morphology reduces the amount of input without disrupting the information structure in the sensorimotor loop. The second result shows how the later morphology under the vergence behavior increases the information structure among the motor actions and the pixels. We also speculate on the implications of our results for attention, reaching and grasping.

Background: One of the main disadvantages in today’s prostheses is their lack of sensory feedback. To overcome this problem, many different ideas have been suggested. One of these approaches is called “sensory substitution”, and exploits the amazing plasticity of our brain. By providing sensory information from one modality through another one, people can regain some of the lost sensory functions. Tactile sensory substitution is achieved by stimulate directly the nerves of mechanoreceptors using a method called transcutaneous electrical nerve stimulation (TENS). In particular, electrical stimulation can be used to generate a moving sensation that we can exploit to transmit dynamical information to the body. Goal / Methods: Healthy subjects were tested on their ability to recognize the direction of the moving sensation. Stimulation patterns could differ in the signal intensity, the stimulation duration, the direction of the moving sensation, and the stimulation side. By analyzing the performance for each factor, we studied the optimal stimulation parameters. Furthermore, we investigated the effect of hemispheric dominance by comparing left-handed and right-handed subjects. Finally, we analyzed the performance in pre- and post-training tests to measure the effect of training on performance. Results / Conclusion: We found that in general, people are able to recognize the moving sensation. The results show that neither handedness, stimulation side, nor stimulation length have an influence on the ability to detect the direction. The important factor is the change in signal intensity required to create the moving sensation. Furthermore, we could show that people benefit from supervised training; they could increase and stabilize their performance. We can conclude that the combination of electrical stimulation and moving sensations provides a promising tool for tactile feedback systems.

null

This work examines correlations between functional morphology and behaviour in the instance of the burrowing locomotion of bivalves. A comparatively simple and assessable behaviour and a rich fossil record documenting the evolutionary adaptations in morphology make these animals adequate for investigation. In this paper a robotic setup to simulate the burrowing behaviour of bivalves is presented. Models of both natural bivalve shell shapes and artificially designed shapes are pulled into sediment in the rocking modality these animals typically use. Different shapes, motion patterns and a water expulsion mechanism are evaluated and compared in terms of burrowing performance. The results presented here and further experiments using the (improved) platform may shed light on how bivalves burrow, how features of functional morphology evolved and how efficient automatic burrowing devices may be constructed. Keywords: biorobotics, biomimetics, underwater robots, functional morphology, burrowing locomotion, shell morphology, bivalves, artificial evolution.

We present a bio-inspired control method for moving-target seeking with a mobile robot, which resembles a predator-prey scenario. The motor repertoire of a simulated Khepera robot was restricted to a discrete number of "gaits". After an exploration phase, the robot automatically synthesizes a model of its motor repertoire, acquiring a forward model. Two additional components were introduced for the task of catching a prey robot. First, an inverse model to the forward model, which is used to determine the action (gait) needed to reach a desired location. Second, while hunting the prey, a model of the prey's behavior is learned online by the hunter robot. All the models are learned ab initio, without assumptions, work in egocentric coordinates, and are probabilistic in nature. Our architecture can be applied to robots with any physical constraints (or embodiment), such as legged robots.

This paper presents the design and control of an actuated bivalve robot, which has been developed to study the burrowing locomotion of bivalves in sediment. The setup consists of a tank filled with sand and water, plastic models of bivalve shells capable of expelling water and an external actuation mechanism simulating the rocking burrowing motion typically used by these animals. The realistic shell shapes have been realized using three-dimensional plotting techniques allowing testing influences of different shell shapes and surface structures (sculptures) on the burrowing efficiency. Based on the experimental setup, the burrowing process has been reproduced. The results show that this setup can be used to identify correlations in the burrowing process. Further experimental work will investigate the influence of factors such as shell shape and sculpture or the motion sequence on the burrowing performance. Keywords: biorobotics; biomimetics; burrowing locomotion; bivalves

With the ability to feel through artificial limbs, users regain more function and increasingly see the prosthetics as parts of their own bodies.

In this thesis, we address a non-invasive sensory feedback system for prosthetic applications. Modern hand prostheses make use of electromyographic (EMG) signals of the remaining hand innervations to control an artificial limb. In contrast to our natural upper limbs, which are sensory-motor coordinated, the communication with a myoelectrically controlled hand prosthesis is only one-directional and consequently no sensory information is fed back to the body. Mainly due to this missing bidirectional information flow, the prosthesis is often not perceived as an integral part of the own body and low user acceptance of prosthesis wearers has been reported. We propose to use transcutaneous electrical stimulation on the lower back to convey pressure as well as proprioceptive feedback regarding the state of a prosthetic hand. For the representation of locally applied force and spatial movements of the hand, we suggest to make use of stationary and moving electrotactile stimulation patterns. We developed a transcutaneous electrical stimulator to investigate to what extent the lower back can be tackled to display sensory feedback. With the developed voltage-controlled device, we performed a series of experiments with 37 volunteer participants. The goal was to assess skin characteristics and basic properties of surface electrical stimulation and to gain insights into the amount of information that could be reliably fed back to the user of a prosthesis. We studied the influence of three different electrode sizes (4.9 cm2, 8 cm2 and 25 cm2) and frequencies (2, 4 and 6 kHz) on the comfort of evoked sensation and on skin characteristics under transcutaneous electrical stimulation. The electrode size and applied frequency did not show a significant effect on perceived stimulation comfort, but on skin impedance, inflowing current, stimulation range, comfort level voltage, current density and power consumption. The applied frequency did not influence the measured parameters to the same extent as electrode size. However, the effect of frequency increased in inverse relation to the size of the electrodes. Larger electrodes showed a reduced skin impedance together with lower required voltages to elicit a sensation and a decreased stimulation range. Furthermore, inter-individual differences were smaller with larger electrodes. Thus, advantages of large electrodes are system stability and low required energy. Drawbacks are the small stimulation range that can be exploited to transmit distinct levels of information and the large covered surface area, which makes stimulation localization more difficult due to a widespread activation area. Consequently, the electrode size represents a trade-off between stability and selectivity and should be determined dependent on the intended application. To gain insights into how much electrotactile information can be reliably recognized on the lower back, we tested the ability of subjects to detect several local and moving stimulation patterns without any previous training. Four electrode pairs were arranged in a rectangular configuration on the lower back and the sensory stimulation was applied either stationary or an up or down traveling sensation between two electrode pairs was induced. We studied the accuracy of sensation detection for two different stimulation waveforms (continuous square pulsed or burst-modulated) and three frequencies (2, 4 and 6 kHz). The mean performance in correctly classifying the applied stimulation patterns was dependent on the stimulation waveform and on the number of simultaneously presented stimuli. The results show an accuracy of unidirectional moving sensation detection of 92% for the best suited stimulation parameters (continuous square pulses). For the best stimulation parameters regarding local stimulation (burst-modulated wave), the mean performance for correct stimulus classification on one electrode pair was 89%. These scores indicate, that local as well as moving sensation patterns can be reliably recognized and the lower back may present a suitable target to display rich and reliable sensory feedback information.

null

In this article, we propose a model which interprets the crossed-hand deficit of temporal order judgment (TOJ) from the dynamical systems perspective. The TOJ paradigm is important to understanding how the body image is dynamically sustained in our daily life. Our aim is to show one possible example of how the TOJ deficit could be consistently expressed in the dynamical systems framework. According to the study reported by Yamamoto and Kitazawa, using the genetic algorithm, we reconstruct the crossed-hand effect with a recurrent neural network, and show that it was caused by slow relaxation dynamics near the critical point in our model agent. Moreover, by using the same agent, we demonstrate the occurrence of the deficit in response to three successive stimuli which may help reveal the mechanism of TOJ deficit in the real human case.

null

null

null

null

null

Cycles is an interactive installation that establishes an intimate relationship between the visitor's physical body and simulated organisms. It explores notions of transience and identity that draw inspiration from Buddhist philosophy. Cycles creates a situation that causes the visitor to experience his or her own body in a state of mutability and transience. Cycles merges the appearance of the visitor's hand with a visual representation of a swarm simulation. By bridging the gap between the virtual and physical, a hybrid entity comes into existence whose rapidly changing body blends artificial and natural properties. This hybrid entity progresses through a life cycle that reenacts the four Buddhist sufferings.

Myoeletric control of robotic prostheses is a widely used approach in medical research. In this work, different algorithms for analyzing the coherence between movement and EMG signals are used and compared. The goal was to find a method which is not only capable of finding coherences between the movement and its directly associated EMG, but also indirectly associated EMG. This aspect is important for later use in upper limb protheses control. An important finding of this work is that EMG signals and movements have a linear, negative or constant coherence, depending on muscle function. This method provides a fast way to compare signals offline. It turned out, that for finding an appropriate method, many factors have to be considered. A challenge for future work is to find a suitable method for analyzing data online. If we know how different movement signals are connected to each other and connected to EMG signals, it would be a big step forward to recognize and predict movements by means of few EMG sensors.