Contributions published at Artificial Intelligence Lab

Animals and humans are constantly faced with a highly dimensional stream of incoming sensory information. At the same time, they have to command their highly complex and multidimensional bodies. Yet, they seamlessly cope with this situation and successfully perform various tasks. For autonomous robots, this poses a challenge: robots performing in the real world are often faced with the curse of dimensionality. In other words, the size of the sensory as well as motor spaces becomes too large for the robot to efficiently cope with them in real time. In this paper, we demonstrate how the curse of dimensionality can be tamed by exploiting the robot’s morphology and interaction with the environment, or the robot’s embodiment (see e.g., [1]). We present three case studies with underactuated quadrupedal robots. In the first case study, we look at terrain detection. While running on different surfaces, the robot generates structured multimodal sensory information that can be used to detect different terrain types. In the second case study, we shift our attention to the motor space: the robot is learning different gaits. The online learning procedure capitalizes on the fact that the robot is underactuated and on a “soft“ control policy. In the third case study, we move one level higher and demonstrate how - given an appropriate gait - a speed adaptation task can be greatly simplified and learned online.

null

null

We introduce a simple model of human’s musculoskeletalsystem to identify the computation that a compliantphysical body can achieve. A one-joint system driven by actuationof the springs around the joint is used as a computationaldevice to compute the temporal integration and nonlinearcombination of an input signal. Only a linear and static readoutunit is needed to extract the output of the computation. Theresults of computer simulations indicate that the network ofmechanically coupled springs can emulate several nonlinearcombinations which need temporal integration. The simulationwith a two-joint system also shows that, thanks to mechanicalconnection between the joints, a distant part of a compliantbody can serve as a computational device driven by the indirectinput. Finally, computational capability of antagonistic musclesand information transfer through mechanical couplings arediscussed.

The development of increasingly complex robots in recent years has been characterized by an extensive use of physics-based simulations for controller design and optimization. Today, a variety of open-source and commercial simulators exist for this purpose for mobile and industrial robots. However, existing simulation engines still lack support for the emerging class of tendon-driven robots. In this paper, an innovative simulation framework for the simulation of tendon-driven robots is presented. It consists of a generic physics simulator capable of utilizing CAD robot models and a set of additional tools for simulation control, data acquisition and system investigation. The framework software architecture has been designed using component-based development principles to facilitate the framework extension and customization. Furthermore, for inter-component communication, the operating-system and programming language independent Common Object Request Broker Architecture (CORBA) [1] has been used which simplifies the integration of the framework into existing software environments.

This article describes an exemplary robot exercise which was conducted in a class for mechatronics students. The goal of this exercise was to engage students in scientific thinking and reasoning, activities which do not always play an important role in their curriculum. The robotic platform presented here is simple in its construction and is customizable to the needs of the teacher. Therefore, it can be used for exercises in many different fields of science, not necessarily related to robotics. Here we present a situation where the robot is used like an alien creature from which we want to understand its behavior, resembling an ethological research activity. This robot exercise is suited for a wide range of courses, from general introduction to science, to hardware oriented lectures.

We present three case studies in which we discuss conceptual and technical aspects of the application of morpholoigcal computation in medical and/or chemical contexts. Up to now, most implementations of morphological computing take profit of classical mechanics and so does one of ours (an inflatable support system for patients with movement impairments). The two other case studies deal with processes and devices on the micrometer scale (self-assembled chemical micro-reactors and models of induced repair in radio-oncology). We use these examples to introduce the notion of embodied process control and discuss how the role taken by classical mechanics in systems on the macro-scale can be adopted by statistical mechanics in case of implementations on the micrometer scale.

The vision of morphological computation proposes that the complexity of compliant bodies of biological systems is not accidentally, but rather that it can contribute to the computations, which are needed for a successful interaction with the environment. We demonstrate in a simulation that a compliant, highly nonlinear body (simulated as a random network of masses and springs) can serve as a computational resource, which allows the end-effector of a two-link robot arm to move autonomously on a complex trajectory. Remarkably, simple linear and static feedback loops from the state of the compliant structure back to the robot arm torques suffice. This suggests that by outsourcing parts of the nonlinear and dynamic computation to the compliant morphology the remaining computational task is much simpler and can be even represented by some static, linear weights.

null

Octopuses have soft and flexible bodies and their behavior is extremely sophisticated. Their behavioral control is accomplished using a simplification strategy, which is based on the division of functionality between the central and peripheral nervous systems. The behavioral control is simplified since the central nervous system only sends the initiation commands to the peripheral nervous system. While the timing to send the command is crucial. In this study, we aim to characterize the relationship between them using a simulated octopus arm. As a result, we show that there exists a clear behavior difference according to the time it takes to initiate the behavior, which may suggest that the behavioral outcome is determined in the early phase of motion generation.

The dynamical coupling of the brain, the body, and the environment is essential to intelligent behaviors. However, we discuss the fact that most current robots still lack this coupling. We propose a methodology to realize such a coupling and demonstrate it in a soft robot experiment platform by releasing the deadtime of the high level controller. Some preliminary results, such as the splitting of the return map and the invariance of average errors, are reported and discussed.

Self-assembly is a phenomenon broadly observed in nature where a vast number of various molecules spontaneously synthesize complex structures. In this paper, aiming at realizing highly autonomous self-assembly systems, we discuss fundamental issues attributed to self-assembly systems that employ magnetism as a driving force. We first introduce some examples from our case studies, in which the models all subscribe to a distributed approach, and thus lack central control. Then we categorize them by their type of magnet attachment. The discussed issues include several fundamental properties, such as the effect of morphology, stochasticity, the difference between 2D models vs. 3D models, emergence, allostericity, and parallelism. The obtained conclusions support our stance, which is that the appropriate morphology lightens the control cost for the assembly, providing primal but engaging instances of magnetic self-assembly systems that warrant further study.

In an embodied agent, the sensor morphology isa fundamental element to shape the information structure ofthe sensorimotor activity. Basically, the density and the sensordistribution work as a filter reducing the dimensionality of thesensory data. The result on a theoretical model shows thatthe discretization is strongly related to the task the agent hasto perform. Moreover, we can take advantage of this relationto define the sensory system, which reduces drastically thedimensionality of the system and highlights the informationstructure in the sensorimotor loop.

null

Crowdsourcing, a real-life instance of human collective intelligence, is a phenomenon that changes the way organizations use the Internet to collect ideas, solve complex cognitive problems, and build high-quality repositories (e.g., Wikipedia) by self-organizing agents around data and knowledge. Many recent studies have highlighted the factors and the small sets of parameters that play a role when a large crowd interacts with an organization. However, no comprehensive simulation has yet been developed to incorporate all these parameters, investigate Artificial Life phenomena such as emergence and self-organization and potentially generate predictive power. Based on a presentation at ALIFE XII, this paper describes the development of a simulator for human crowds performing collective problem solving in a Crowdsourcing scenario. It introduces the mechanics of a multi-agent system (MAS) by building on insights from empirical science in several disciplines. The simulator allows running sensitivity analyses of multiple parameters as well as simulation of intractable interactions of complex networks of irrational agents. In addition, the paper provides a review of Crowdsourcing and human collective intelligence literature structured from an Alife point-of-view.

Students today are a generation of consumers of technological devices and software. They tend to consider themselves as technology savvy and digital natives. However, this pervasive use of technology seems to encourage mastering the usage of a device rather than the interest to understand the technical details behind it. This might help to explain the decreasing number of enrollments in science and technology disciplines at Universities in the USA as well as in Europe. In this paper we introduce our concept of promoting interest in science and technology through understanding technological gadgets young people are familiar with. We introduce an implementation of this concept, an open embedded systems kit for education (EmbedIT) which currently is under development. Unlike common educational robot kits EmbedIT enables students to access the technical world in a non-engineering focused way. Through a graphical user interface students can play with sensors and actuators and are able to easily implement technological objects using them. 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.

The aim of this study is to explore a control architecture that can control a soft and "exible 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.

null

null