Teaching motion skills to robots through demonstrations has becomes widely popular. However, precise execution of start-, via-, and end-poses at given times is often not guaranteed, limiting the technology transfer to industrial application. To address this issue, we propose the novel Constrained Expectation Maximization (CEM) algorithm, which enforces time-sensitive constraints (TSC) when learning Gaussian Mixture Models (GMM). Our approach applies to data on Riemannian manifolds and extends to task-parameterized scenarios. We validate CEM against state-of-the-art methods on handwritten data and real robot applications utilizing the KUKA LBR iiwa. By enforcing constraints within the learning process, CEM achieves improved and more efficient reproduction of the demonstration data.

Temporal alignment of multiple signals through time warping is crucial in many fields, such as classification within speech recognition or robot motion learning. Almost all related works are limited to data in Euclidean space. Although an attempt was made in 2011 to adapt this concept to unit quaternions, a general extension to Riemannian manifolds remains absent. Given its importance for numerous applications in robotics and beyond, we introduce Riemannian Time Warping (RTW). This novel approach efficiently aligns multiple signals by considering the geometric structure of the Riemannian manifold in which the data is embedded. Extensive experiments on synthetic and real-world data, including tests with an LBR iiwa robot, demonstrate that RTW consistently outperforms state-of-the-art baselines in both averaging and classification tasks.

The robot-based tracking of highly dynamic end point motions of deformable linear objects (DLO) remains challenging due to its non-linear behavior. Since simple feedback control is infeasible, model-based control offers potential to account for the non-linear effects, but requires computation efficient and accurate models. Promising results have been achieved utilizing data-driven models that introduce a latent kinematic chain as model of the DLO and mapping measurements of the tip position in its latent joint space, in which the dynamic motion model is learned. So far, this approach has the limitation that it can not handle situations of incomplete sensory information, for instance if occlusion occurs. Consequently, this paper introduces a fusion network architecture capable of making predictions even if sensory information is incomplete. We achieve additional state estimation of the latent joint state by learning a data driven inverse kinematics with help of wrench measurements at the DLO base and evaluate our approach by simulating occlusion. We demonstrate the computational effectiveness of our approach for in the loop control tasks.

Data-driven techniques show promising results in force estimation for Concentric Tube Continuum Robots, but often require extensive datasets, which are difficult to acquire. This work introduces a mapping-based transfer learning approach to improve the efficiency of data-driven methods for contact force estimation, by proposing a diffeomorphic mapping-based algorithm that reduces data requirements, enabling more practical application of these methods. By transforming data from pre-curved tubes into the feature space of non-curved tubes, our method allows a pre-trained neural network to estimate forces efficiently across various tube configurations, eliminating the need for additional data for changing tube configurations and retraining of the network. Simulation tests show high accuracy for curvatures up to kappa = 7 1/m, significantly reducing the need to create large datasets for each new robot configuration.

This work presents a novel approach to handling epistemic uncertainty estimates with motivation from Bayesian linear regression. We propose treating the model-dependent variance in the predictive distribution-commonly associated with epistemic uncertainty-as a model for the underlying data distribution. Using high-dimensional random feature transformations, this approach allows for a computationally efficient, parameter-free representation of arbitrary data distributions. This allows assessing whether a query point lies within the distribution, which can also provide insights into outlier detection and generalization tasks. Furthermore, given an initial input, minimizing the uncertainty using gradient descent offers a new method of querying data points that are close to the initial input and belong to the distribution resembling the training data, much like auto-completion in associative networks. We extend the proposed method to applications such as local Gaussian approximations, input-output regression, and even a mechanism for unlearning of data. This reinterpretation of uncertainty, alongside the geometric insights it provides, offers an innovative and novel framework for addressing classical machine learning challenges.

The search for ethical guidance in the development of artificial intelligence (AI) systems, especially in healthcare and decision support, remains a crucial effort. So far, principles usually serve as the main reference points to achieve ethically correct implementations. Based on reviewing classical criticism of principle-based ethics and taking into account the severity and potentially life-changing relevance of decisions assisted by AI-driven systems, we argue for strengthening a complementary perspective that focuses on the life-world as ensembles of practices which shape people's lives. This perspective focuses on the notion of ethical judgment sensitive to life forms, arguing that principles alone do not guarantee ethicality in a moral world that is rather a joint construction of reality than a matter of mere control. We conclude that it is essential to support and supplement the implementation of moral principles in the development of AI systems for decision-making in healthcare by recognizing the normative relevance of life forms and practices in ethical judgment.

Assistive devices like exoskeletons undergo extensive testing not least because of their close interaction with humans. Conducting user studies is a time-consuming process that demands expert knowledge, and it is accompanied by challenges such as low repeatability and a potential lack of comparability between studies. Obtaining objective feedback on the exoskeleton’s performance is crucial for developers and manufacturers to iteratively improve the design and development process. This paper contributes to the concept of using robots for objective exoskeleton testing by presenting various approaches to a robotic-based testing platform for upper-body exoskeletons. We outline the necessary requirements for realistically simulating use cases and evaluate different approaches using standard manipulators as robotic motion generators. Three approaches are investigated: (i) Exploiting the anthropomorphic structure of the robotic arm and directly placing it into the exoskeleton. (ii) Utilizing a customized, direct attachment between the robot and exoskeleton. (iii) Attaching a human arm dummy to the robot end effector to simulate a more realistic interface with the exoskeleton. Subsequently, we discuss and compare the results against the aforementioned requirements of a systematic testing platform. Our conclusion emphasizes that achieving objective and realistic testing necessitates highly specialized hardware, algorithms, and further research to address challenging requirements.

Programming a robot manipulator should be as intuitive as possible. To achieve that, the paradigm of teaching motion skills by providing few demonstrations has become widely popular in recent years. Probabilistic versions thereof take into account the uncertainty given by the distribution of the training data. However, precise execution of start-, via-, and end-poses at given times can not always be guaranteed. This limits the technology transfer to industrial application. To address this problem, we propose a novel constrained formulation of the Expectation Maximization algorithm for learning Gaussian Mixture Models (GMM) on Riemannian Manifolds. Our approach applies to probabilistic imitation learning and extends also to the well-established TP-GMM framework with Task-Parameterization. It allows to prescribe end-effector poses at defined execution times, for instance for precise pick & place scenarios. The probabilistic approach is compared with state-of-the-art learning-from-demonstration methods using the KUKA LBR iiwa robot. The reader is encouraged to watch the accompanying video available at https://youtu.be/JMI1YxtN9C0

Concentric tube continuum robots (CTCRs) belong to the family of continuum robots with applications in minimally invasive surgeries. Because of this application domain, measuring the external forces along the body of the robot is paramount. CTCRs are made up of thin elastic rods and are intended to be applied inside the human body, where conventional sensor-based measurements are not feasible. Consequently, research is resorting to estimate the forces through geometric, numeric, or optimization methods. However, these methods often suffer from slow convergence. In this paper, we introduce a novel data-driven approach for estimating contact forces along the body of a CTCR that offers an estimation precision comparable to the current state-of-the-art optimization-based approaches, but exhibits nearly two orders of magnitude faster convergence. The proposed method is scalable and exhibits a significant performance in response to a wide range of external forces. The approach was evaluated in simulations and on a real 2-tube CTCR.

It has been emphasized for a long time that real-world applications of developmental robots require lifelong and online learning. A major challenge in this field is the high sample-complexity of algorithms, which has led to the development of intrinsic motivation approaches to render learning more efficient. However, only few works have been demonstrated on real robots and although these robots are supposed to share the environment with humans, there is hardly any research to integrate intrinsic motivation with learning from an interacting teacher. In this article, we tackle the efficiency challenge by proposing a novel extrinsic–intrinsic motivation learning scheme. We specifically investigate how to combine intrinsic motivation with learning from observation to accelerate learning. Our novel scheme comprises four elements: 1) a probabilistic intrinsic motivation signal yielding the robot’s interest; 2) a probabilistic extrinsic motivation signal to expand the robot’s knowledge by learning from observation; 3) novelty detection; and 4) novelty degree methods to enable the robot to decide autonomously how and when to explore. The efficiency as well as the applicability of our methods are benchmarked in simulation experiments and demonstrated on a physical 7-degree of freedom left arm of Baxter robot.

Robotics research into multi-robot systems so far has concentrated on implementing intelligent swarm behavior and contact-less human interaction. Studies of haptic or physical human-robot interaction, by contrast, have primarily focused on the assistance offered by a single robot. Consequently, our understanding of the physical interaction and the implicit communication through contact forces between a human and a team of multiple collaborative robots is limited. We here introduce the term Physical Human Multi-Robot Collaboration (PHMRC) to describe this more complex situation, which we consider highly relevant in future service robotics. The scenario discussed in this article covers multiple manipulators in close proximity and coupled through physical contacts. We represent this set of robots as fingers of an up-scaled agile robot hand. This perspective enables us to employ model-based grasping theory to deal with multi-contact situations. Our torque-control approach integrates dexterous multi-manipulator grasping skills, optimization of contact forces, compensation of object dynamics, and advanced impedance regulation into a coherent compliant control scheme. For this to achieve, we contribute fundamental theoretical improvements. Finally, experiments with up to four collaborative KUKA LWR IV+ manipulators performed both in simulation and real world validate the model-based control approach. As a side effect, we notice that our multi-manipulator control framework applies identically to multi-legged systems, and we execute it also on the quadruped ANYmal subject to non-coplanar contacts and human interaction.

Objective This meta-analysis reviews robot design features of interface, controller, and appearance and statistically summarizes their effect on successful human–robot interaction (HRI) at work (that is, task performance, cooperation, satisfaction, acceptance, trust, mental workload, and situation awareness). Background Robots are becoming an integral part of many workplaces. As interactions with employees increase, ensuring success becomes ever more vital. Even though many studies investigated robot design features, an overview on general and specific effects is missing. Method Systematic selection of literature and structured coding led to 81 included experimental studies containing 380 effect sizes. Mean effects were calculated using a three-level meta-analysis to handle dependencies of multiple effect sizes in one study. Results Sufficient feedback through the interface, clear visibility of affordances, and adaptability and autonomy of the controller significantly affect successful HRI, whereas appearance does not. The features of the interface and controller affect performance and satisfaction but do not affect situation awareness and trust. Specific effects of adaptability on cooperation and acceptance, as well as autonomy on mental workload, could be shown. Conclusion Robot design at work needs to cover multiple features of interface and controller to achieve successful HRI that covers not only performance and satisfaction, but also cooperation, acceptance, and mental workload. More empirical research is needed to investigate mediating mechanisms and underrepresented design features’ effects. Application Robot designers should carefully choose design features to balance specific effects and implementation costs with regard to tasks, work design aims, and employee needs in the specific work context.

Due to the change of industrial processes and demographic shift in many countries, an increase in the use and application of exoskeletons is expected. However, design, development and deployment of exoskeletons requires testing. The standard way of testing novel interactive technologies by user studies suffers from a number of limitations, namely low repeatability between and within subjects, high resource demands, and technical and logistical issues. In this paper, we therefore promote the idea to use robots for more systematic and effective testing of exoskeletons. We suggest a threefold methodology that (i) employs models and simulation to investigate essential mechanisms of the human’s or the robot’s interaction with the exoskeleton, (ii) relies on capturing human motion and using machine learning to model it, and (iii) develops a dedicated platform to deploy the learned human-like motion models on the testing robot. In particular, we argue for a robot design that is specifically tailored to this testing task and can exhibit human-like motion. In this paper we discuss our methodology and its steps towards design and development of such a dedicated test platform.

A major challenge for online and data-driven model learning in robotics is the high sample complexity. This hinders its efficiency and practical feasibility for lifelong learning, in particular, for developmental robots that autonomously bootstrap their sensorimotor skills in an open-ended environment. In this work, we propose new methods to mediate this problem in order to permit the learning of robot models online, from scratch, and in learning while behaving fashion. Exploration is utilized and autonomously driven by a novel intrinsic motivation signal which combines knowledge-based and competence-based elements and surpasses other state-of-the-art methods. In addition, we propose an episodic online mental replay to accelerate online learning, to ensure sample efficiency, and to update the model online rapidly. The efficiency as well as the applicability of our methods are demonstrated with a physical 7-DoF Baxter manipulator. We show that our learning schemes are able to drastically reduce the sample complexity and learn the data-driven model online, even within a limited time frame.

Humans perform dexterous and highly dynamic movements like throwing, running, hitting or jumping by adjusting the dynamic characteristics of their musculoskeletal system. While the human muscle stiffness varies across different tasks, adjustable stiffness ensures the safety, saves energy, and provides comfort of to the human while maintaining performance. Roboticists try to integrate similar capabilities into robotic manipulators by using dedicated actuators such as Variable Stiffness Actuators (VSAs). VSAs integrate elastic elements with either fixed or variable mechanical compliance to resemble human muscles. The usage of such compliance elements also allows to actively control the stiffness. However, the full potential of such actuation has currently not been realized because it is notoriously difficult to control VSAs, specifically in the low-stiffness regime which, however, is particularly safe and also most suited to store energy for highly dynamic motions release. In this paper, we therefore consider control of an VSA manufactured by qbrobotics. The qbmove VSA consists of two motors which connect to the output shaft by nonlinear elastic elements such that both motors contribute torques symmetrically to produce shaft rotation, while antagonistic motor actuation regulates the stiffness. We realize and evaluate to this aim conventional PID control, Sliding Mode Control (SMC) and Model Predictive Control (MPC). Based on a dynamical model of the VSA their performance is evaluated for tracking a sinusoidal and a square wave. To apply linear MPC, an approximate linear VSA model is created through system identification of the actual hardware and likewise applied. The results show superior performance of the MPC as compared to the standard PID baseline and Sliding Mode Control.

This paper proposes a novel approach to automatically generate labeled training data for predicting parallel-jaw grasps from stereo-matched depth images. We generate realistic depth images using Semi-Global Matching to compute disparity maps from synthetic data, which allows producing images that mimic the typical artifacts from real stereo matching in our data, thus reducing the gap from simulation to real execution. Our pipeline automatically generates grasp annotations for single or multiple objects on the synthetically rendered scenes, avoiding any manual image pre-processing steps such as inpainting or denoising. The labeled data is then used to train a CNN-model that predicts parallel-jaw grasps, even in scenarios with large amount of unknown depth values. We further show that scene properties such as the presence of obstacles (a bin, for instance) can be added to our pipeline, and the training process results in grasp prediction success rates of up to 90%.

Shape-sensing in real-time is a key requirement for the development of advanced algorithms for concentric tube continuum robots when safe interaction with the environment is important e.g., for path planning, advanced control, and human-machine interaction. We propose a real-time shape-estimation algorithm for concentric tube continuum robots based on the force-torque information measured at the tubes’ basis. It extends a shape estimation algorithm for elastic rods based on discrete Kirchhoff rod theory. For simplicity and efficiency of calculation, we combine it with a model under piece-wise constant curvature assumption, in which we model a concentric tube continuum robot as a combination of segments of planar constant curvatures lying on different equilibrium planes. We evaluate our approach for a single and two combined additively manufactured tubes and achieve an estimation frequency of 333 Hz for two combined tubes with a mean deviation along the backbone of the tubes of 1.91–5.22 mm.

We propose a novel approach that biases actions during policy search by lifting the concept of redundancy resolution from multi-DoF robot kinematics to the level of the reward in deep reinforcement learning and evolution strategies. The key idea is to bias the distribution of executed actions in the sense that the immediate reward remains unchanged. The resulting biased actions favor secondary objectives yielding policies that are safer to apply on the real robot. We demonstrate the feasibility of our method, considered as policy search with redundant action bias (PSRAB), in a reaching and a pick-andlift task with a 7-DoF Franka robot arm trained in RLBench - a recently introduced benchmark for robotic manipulation - using state-of-the-art TD3 deep reinforcement learning and OpenAI's evolutionary strategy. We show that it is a flexible approach without the need of significant fine-tuning and interference with the main objective even across different policy search methods and tasks of different complexity. We evaluate our approach in simulation and on the real robot. Our project website with videos and further results can be found at: https://sites.google.com/view/redundant-action-bias

Zusammenfassung Neue technologische Möglichkeiten und künstliche Intelligenz ermöglichen in der Mensch-Technik Interaktion immer neue und zunehmend anspruchsvollere Assistenzfunktionen. Dieser Beitrag widmet sich dabei den Fragen, welche Assistenz und wieviel Assistenz sinnvoll ist. Die aktuelle Forschung zeigt, dass die Beantwortung dieser Fragen nicht trivial ist, denn technische Assistenz muss kognitiv angemessen und in der Interaktion sinnvoll eingebettet sein, um Nutzern tatsächlich bei der Erfüllung ihrer Aufgaben zu helfen. Im vorliegenden Buchkapitel liegt daher der Fokus auf der Analyse und der kritischen Reflexion genau dieser Interaktion zwischen menschlicher und künstlicher Intelligenz. Es reflektiert die Rollenverteilungen und die Wirkungen auf den jeweils anderen Interaktionspartner und zeigt auf, welche neuen Forschungsfragen in diesem Kontext diskutiert und beantwortet werden müssen, um ein sich erweiterndes Zusammenwirken zwischen Mensch und Technik gewinnbringend zu gestalten.