Contributions published at Artificial Intelligence Lab

This study aims to explore a control architecture that enables the control of a soft and flexible octopus-like arm for an object reaching task. Inspired by the division of functionality between the central and peripheral nervous systems of a real octopus, we discuss that the important factor of the control is not to regulate the arm muscles one by one but rather to control them globally with appropriate timing, and we propose an architecture equipped with a recurrent neural network (RNN). By setting the task environment for the reaching behavior, and training the network with an incremental learning strategy, we evaluate whether the network is then able to achieve the reaching behavior or not. As a result, we show that the RNN can successfully achieve the reaching behavior, exploiting the physical dynamics of the arm due to the timing based control.

Robots have often been used as an educational tool in class to introduce kids to science and technology, disciplines that are affected by decreasing enrollments in universities. Consequently, many robotic kits are available off-the-shelf. Even though many of these platforms are easy to use, they focus on a classical top-down engineering approach. Additionally, they often require advanced programming skills. In this paper we introduce an open robotic kit for education (EmbedIT) which currently is under development. Unlike common robot kits EmbedIT enables students to access the technical world in a non-engineering focused way. Through a graphical user interface students can easily build and control robots. We believe that once fascination and a basic understanding of technology has been established, the barrier to learn more advanced topics such as programming and electronics is lowered. Further we describe the hardware and software of EmbedIT, the current state of implementation, and possible applications.

Compliant actuation contributes enormously in legged locomotion robotics since it is able to alleviate control efforts in improving the robot’s adaptability and energy efficiency. In this paper, we present a novel design of a variable stiffness rotary actuator, called MESTRAN, which was especially targeted to address the limitations in terms of the amount of energy and time required to vary the stiffness of an actuated joint. We have constructed a mechanical model in simulation and a physical prototype. We conducted a series of experiments to validate the performance of the MESTRAN actuator prototype. The results from the simulation and experiments show that MESTRAN allows independent control of stiffness and position of an actuated rotary joint with a large operational range and high speed. The torque-displacement relationship is close to linear. Lastly, the MESTRAN actuator is energy-efficient since a certain stiffness level is maintained without energy input.

It is an important ability for any mobile robot to be able to estimate its posture and to gauge the distance it travelled. The information can be obtained from various sources. In this work, we have addressed this problem in a dynamic quadruped robot. We have designed and implemented a navigation algorithm for full body state (position, velocity, and attitude) estimation that does not use any external reference (such as GPS, or visual landmarks). Extended Kalman Filter was used to provide error estimation and data fusion from two independent sources of information: Inertial Navigation System mechanization algorithm processing raw inertial data, and legged odometry, which provided velocity aiding. We present a novel data-driven architecture for legged odometry that relies on a combination of joint sensor signals and pressure sensors. Our navigation system ensures precise tracking of a running robot's posture (roll and pitch), and satisfactory tracking of its position over medium time intervals. We have shown our method to work for two different dynamic turning gaits and on two terrains with significantly different friction. We have also successfully demonstrated how our method generalizes to different velocities.

Fish excel in their swimming capabilities. These result from a dynamic interplay of actuation, passive properties of fish body, and interaction with the surrounding fluid. In particular, fish are able to exploit wakes that are generated by objects in flowing water. A powerful demonstration that this is largely due to passive body properties are studies on dead trout. Inspired by that, we developed a multi joint swimming platform that explores the potential of a passive dynamic mechanism. The platform has one actuated joint only, followed by three passive joints whose stiffness can be changed online, individually, and can be set to an almost arbitrary nonlinear stiffness profile. In a set of experiments, using online optimization, we investigated how the platform can discover optimal stiffness distribution along its body in response to different frequency and amplitude of actuation. We show that a heterogeneous stiffness distribution - each joint having a different value - outperforms a homogeneous one in producing thrust. Furthermore, different gaits emerged in different settings of the actuated joint. This work illustrates the potential of online adaption of passive body properties, leading to optimized swimming, especially in an unsteady environment.

Biomimetics is a fruitful combination of biology and engineering, leading not only to technological innovations but also new nsights into biological questions. In this ongoing project, embodied artificial intelligence (embodied AI), artificial evolution and palaeontology are combined to investigate the functional morphology of bivalves. This cross-fertilization allows to expand biomimetics from current biological systems to the whole evolutionary history and to apply the synthetic approach common in embodied AI as a method to tackle open palaeontological questions. So far, a robotic platform has been built to mimic the burrowing technique applied by bivalves. First results show interesting insights into underwater burrowing. We plan to build a more complex version of the system and to perform evolutionary robotics experiments.

Scaling laws are pervasive in biological systems, found in a large number of life processes, and across 27 orders of magnitude. Recent findings show both biological and engineered motors adhering to two fundamental regimes for the mass scaling of maximum force output. This scaling law is of particular interest for the robotics field as it can affect the design stage of a robot. In this study we present data of motors commonly used in robotic applications and find an adherence to a similar power law of mass scaling of maximum torque output in two groups, group a, (Ga ∝ m1.00) and group b (Gb ∝ m1.27). Findings imply that there could exist an upper motor limit of maximum specific torque/force that should be taken under consideration in robot design. Additionally, we show how a robot's minimum mass can be calculated with motor mass being the only necessary parameter.

The aim of this study is to explore a control architecture that facilitates the control of a soft and flexible octopus-like arm. Inspired by the division of functionality between the central and peripheral nervous systems of a real octopus, we note that the important requirement for control is not to regulate the arm muscles one by one but rather to control them collectively with the appropriate timing. In order to realize this timing-based control, we propose an architecture that is equipped with a recurrent neural network (RNN) and then we determine the performance of its reaching behavior. To train the network, we introduce an incremental learning strategy that is capable of taking the body's dynamics into account. As a result, we show that the RNN can successfully accomplish the reaching behavior by exploiting the physical dynamics of the arm due to the timing-based control.

The ShanghAI Lectures project contributes to the fundamental goal of making education and knowledge accessible to a broad interdisciplinary and intercultural audience. Deploying state-of-the-art videoconferencing technology and three-dimensional virtual environments, the project enables students and researchers from all around the globe to learn and work together.

null

The problem of movement coordination in large DoF (Degree of Freedom) robots is complex due to redundancies. In this regard, Dynamic Movement Primitive (DMP) is a useful planning technique, inspired by biology, that can be used to store and reproduce trajectories about every DoF. This work is a preliminary study that aims to understand and quantify the influence of the robot dynamics upon the performance of DMP in a simulated 2DoF robot arm. The investigation demonstrates that the effect of the robot body dynamics needs to be taken into account during the learning process of the DMP.

This paper describes two robot competitions that took place within a robotics class for teachers. The robotics class was part of a two-year master's degree program that aims at educating upper secondary school teachers of different backgrounds in informatics, a discipline that is not yet a mandatory part of the school curriculum in Switzerland. The aim of these robot competitions was to familiarize the teachers with robotic hardware and software such that they would be able to design their own informatics class syllabus. We describe the robotic platforms used, the competitions, their aims and results. Furthermore, we address the question whether robots are a suitable tool for teacher education in informatics.

The control of compliant robots is, due to their often nonlinear and complex dynamics, inherently difficult. The vision of morphological computation proposes to view these aspects not only as problems, but rather also as parts of the solution. Non-rigid body parts are not seen anymore as imperfect realizations of rigid body parts, but rather as potential computational resources. The applicability of this vision has already been demonstrated for a variety of complex robot control problems. Nevertheless, a theoretical basis for understanding the capabilities and limitations of morphological computation has been missing so far. We present a model for morphological computation with compliant bodies, where a precise mathematical characterization of the potential computational contribution of a complex physical body is feasible. The theory suggests that complexity and nonlinearity, typically unwanted properties of robots, are desired features in order to provide computational power. We demonstrate that simple generic models of physical bodies, based on mass-spring systems, can be used to implement complex nonlinear operators. By adding a simple readout (which is static and linear) to the morphology such devices are able to emulate complex mappings of input to output streams in continuous time. Hence, by outsourcing parts of the computation to the physical body, the difficult problem of learning to control a complex body, could be reduced to a simple and perspicuous learning task, which can not get stuck in local minima of an error function.

In this thesis we investigate how the social organization of fish schools is influenced by the morphology and the sensory capabilities of the individuals as well as by those of their predators. We do this by means of individual-based models. Here, behavior at the group-level (schooling) is a consequence of local interactions, i.e. the responses of individuals to their neighbors and the interactions between predator and prey.We demonstrate how modeling the embodiment and the perceptual capabilities (situatedness) both of the individuals and of the predator influences their interaction and therefore the patterns at the group-level. Representing the individuals’ body affects the inter-individual spacing, such that large individuals occupy more space compared to small ones. Modeling the individuals’ situatedness, by reflecting the masking of distant neighbors by closer ones, restricts interaction to the local environment of the individual. This influences many schooling characteristics, such as nearest neighbor distance or group speed, and in mixed schools of large and small individuals it leads to the segregation of the two sizes. In large groups school shape becomes complex and variable and the distribution of individuals heterogeneous, with regions of high and low density occurring anywhere in the school. Modeling morphological and sensory constraints of a predator affects its success in capturing prey and, therefore, influences whether schooling behavior is beneficial for the individuals or not. We demonstrate that when the predator is confusable, i.e. when its sensory capabilities to detect the movements of individuals in a group are limited, schooling is almost always beneficial. In summary, incorporating aspects of embodiment and situatedness leads to more realistic models, first, because the real world is reflected more accurately, and, second, because they lead to a more realistic social organization of the simulated schools.

This master’s thesis deals with the implementation of a motion capture system measuring joints’ angles and estimating the kinematic model of human and the ECCERobot in real time as well as of a control mechanism for robot’s ‘muscles’. In the framework of my master project I created several components. The first component is used for motion capture from a human. For this purpose, I used a Kinect camera. The module was built in a way to control the camera, visualize a human 3D character from the data received and estimate the position of her limbs. The second component has the same functionality, but it was created for the robot. Kinect could not ‘catch’ the robot’s skeleton, so this I used retro-reflective markers, OptiTrack cameras and Tracking Tools software. I suggested, developed and tested a method of estimation of the robot’s upper body kinematical model. The method showed accurate and stable results. Also I proposed an algorithm based on an echo state neural network for the reproduction of human movement and control of the robot’s 'muscles'.

null

null