Most recent successes in robot reinforcement learning involve learning a specialized single-task agent. However, robots capable of performing multiple tasks can be much more valuable in real-world applications. Multi-task reinforcement learning can be very challenging due to the increased sample complexity and the potentially conflicting task objectives. Previous work on this topic is dominated by model-free approaches. The latter can be very sample inefficient even when learning specialized single-task agents. In this work, we focus on model-based multi-task reinforcement learning. We propose a method for learning multi-task visual world models, leveraging pre-trained language models to extract semantically meaningful task representations. These representations are used by the world model and policy to reason about task similarity in dynamics and behavior. Our results highlight the benefits of using language-driven task representations for world models and a clear advantage of model-based multi-task learning over the more common model-free paradigm.

This study presents an innovative test rig engineered to explore the kinematic and viscoelastic characteristics of human specimen hands. The rig features eight force-controlled motors linked to muscle tendons, enabling precise stimulation of hand specimens. Hand movements are monitored through an optical tracking system, while a force-torque sensor quantifies the resultant fingertip loads. Employing this setup, we successfully demonstrated a pincer grip using a cadaver hand and measured both muscle forces and grip strength. Our results reveal a nonlinear relationship between tendon forces and grip strength, which can be modeled by an exponential fit. This investigation serves as a nexus between biomechanical and robotics-focused research, providing critical insights for the advancement of robotic hand actuation and therapeutic interventions.

In robotics and biomechanics, accurately determining joint parameters and computing the corresponding forward and inverse kinematics are critical yet often challenging tasks, especially when dealing with highly individualized and partly unknown systems. This paper unveils a cutting-edge kinematic optimizer, underpinned by an autoencoder-based architecture, to address these challenges. Utilizing a neural network, our approach simulates inverse kinematics, converting measurement data into joint-specific parameters during encoding, enabling a stable optimization process. These parameters are subsequently processed through a predefined, differentiable forward kinematics model, resulting in a decoded representation of the original data. Beyond offering a comprehensive solution to kinematics challenges, our method also unveils previously unidentified joint parameters. Real experimental data from knee and hand joints validate the optimizer’s efficacy. Additionally, our optimizer is multifunctional: it streamlines the modeling and automation of kinematics and enables a nuanced evaluation of diverse modeling techniques. By assessing the differences in reconstruction losses, we illuminate the merits of each approach. Collectively, this preliminary study signifies advancements in kinematic optimization, with potential applications spanning both biomechanics and robotics.

In offline reinforcement learning, a policy is learned using a static dataset in the absence of costly feedback from the environment. In contrast to the online setting, only using static datasets poses additional challenges, such as policies generating out-of-distribution samples. Model-based offline reinforcement learning methods try to overcome these by learning a model of the underlying dynamics of the environment and using it to guide policy search. It is beneficial but, with limited datasets, errors in the model and the issue of value overestimation among out-of-distribution states can worsen performance. Current model-based methods apply some notion of conservatism to the Bellman update, often implemented using uncertainty estimation derived from model ensembles. In this paper, we propose Constrained Latent Action Policies (C-LAP) which learns a generative model of the joint distribution of observations and actions. We cast policy learning as a constrained objective to always stay within the support of the latent action distribution, and use the generative capabilities of the model to impose an implicit constraint on the generated actions. Thereby eliminating the need to use additional uncertainty penalties on the Bellman update and significantly decreasing the number of gradient steps required to learn a policy. We empirically evaluate C-LAP on the D4RL and V-D4RL benchmark, and show that C-LAP is competitive to state-of-the-art methods, especially outperforming on datasets with visual observations.

Acknowledging that society is made up of different sectors with their own rules and structures, this paper studies the relevance of a sector-specific perspective to AI ethics. Incidents with AI are studied in relation to five sectors (police, healthcare, education and academia, politics, automotive) using the AIAAIC repository. A total of 125 incidents are sampled and analyzed by conducting a qualitative content analysis on media reports. The results show that certain ethical principles are found breached across sectors: accuracy/reliability, bias/discrimination, transparency, surveillance/privacy, security. However, results also show that 1) some ethical issues (misinformation, safety, premise/intent) are sector specific, 2) the consequences and meaning of the same ethical issue is able to vary across sectors and 3) pre-existing sector-specific issues are reproduced with these ethical breaches. The paper concludes that general ethical principles are relevant to discuss across sectors, yet, a sector-based approach to AI ethics gives in-depth information on sector-specific structural issues.

Unlike most reinforcement learning agents which require an unrealistic amount of environment interactions to learn a new behaviour, humans excel at learning quickly by merely observing and imitating others. This ability highly depends on the fact that humans have a model of their own embodiment that allows them to infer the most likely actions that led to the observed behaviour. In this paper, we propose Action Inference by Maximising Evidence (AIME) to replicate this behaviour using world models. AIME consists of two distinct phases. In the first phase, the agent learns a world model from its past experience to understand its own body by maximising the ELBO. While in the second phase, the agent is given some observation-only demonstrations of an expert performing a novel task and tries to imitate the expert's behaviour. AIME achieves this by defining a policy as an inference model and maximising the evidence of the demonstration under the policy and world model. Our method is "zero-shot" in the sense that it does not require further training for the world model or online interactions with the environment after given the demonstration. We empirically validate the zero-shot imitation performance of our method on the Walker and Cheetah embodiment of the DeepMind Control Suite and find it outperforms the state-of-the-art baselines. Code is available at: https://github.com/argmax-ai/aime.

Placing robotsoutside controlled conditions requires versatile movement representations that allow robots to learn new tasks and adapt them to environmental changes. The introduction of obstacles or the placement of additional robots in the workspace and the modification of the joint range due to faults or range-of-motion constraints are typical cases, where the adaptation capabilities play a key role for safely performing the robot’s task. Probabilistic movement primitives (ProMPs) have been proposed for representing adaptable movement skills, which are modeled as Gaussian distributions over trajectories. These are analytically tractable and can be learned from a small number of demonstrations. However, both the original ProMP formulation and the subsequent approaches only provide solutions to specific movement adaptation problems, e.g., obstacle avoidance, and a generic, unifying, probabilistic approach to adaptation is missing. In this article, we develop a generic probabilistic framework for adapting ProMPs. We unify previous adaptation techniques, for example, various types of obstacle avoidance, via-points, and mutual avoidance, in one single framework and combine them to solve complex robotic problems. Additionally, we derive novel adaptation techniques such as temporally unbound via-points and mutual avoidance. We formulate adaptation as a constrained optimization problem, where we minimize the Kullback–Leibler divergence between the adapted distribution and the distribution of the original primitive,while we constrain the probability mass associated with undesired trajectories to be low. We demonstrate our approach on several adaptation problems on simulated planar robot arms and seven-degree-of-freedom Franka Emika robots in a dual-robot-arm setting.

Deep state-space models (DSSMs) enable temporal predictions by learning the underlying dynamics of observed sequence data. They are often trained by maximising the evidence lower bound. However, as we show, this does not ensure the model actually learns the underlying dynamics. We therefore propose a constrained optimisation framework as a general approach for training DSSMs. Building upon this, we introduce the extended Kalman VAE (EKVAE), which combines amortised variational inference with classic Bayesian filtering/smoothing to model dynamics more accurately than RNN-based DSSMs. Our results show that the constrained optimisation framework significantly improves system identification and prediction accuracy on the example of established state-of-the-art DSSMs. The EKVAE outperforms previous models w.r.t. prediction accuracy, achieves remarkable results in identifying dynamical systems, and can furthermore successfully learn state-space representations where static and dynamic features are disentangled.

Measuring the similarity between data points often requires domain knowledge, which can in parts be compensated by relying on unsupervised methods such as latent-variable models, where similarity/distance is estimated in a more compact latent space. Prevalent is the use of the Euclidean metric, which has the drawback of ignoring information about similarity of data stored in the decoder, as captured by the framework of Riemannian geometry. We propose an extension to the framework of variational auto-encoders allows learning flat latent manifolds, where the Euclidean metric is a proxy for the similarity between data points. This is achieved by defining the latent space as a Riemannian manifold and by regularising the metric tensor to be a scaled identity matrix. Additionally, we replace the compact prior typically used in variational auto-encoders with a recently presented, more expressive hierarchical one—and formulate the learning problem as a constrained optimisation problem. We evaluate our method on a range of data-sets, including a video-tracking benchmark, where the performance of our unsupervised approach nears that of state-of-the-art supervised approaches, while retaining the computational efficiency of straight-line-based approaches.

Learning from multiple sources of information is an important problem in machine-learning research. The key challenges are learning representations and formulating inference methods that take into account the complementarity and redundancy of various information sources. In this paper we formulate a variational autoencoder based multi-source learning framework in which each encoder is conditioned on a different information source. This allows us to relate the sources via the shared latent variables by computing divergence measures between individual source’s posterior approximations. We explore a variety of options to learn these encoders and to integrate the beliefs they compute into a consistent posterior approximation. We visualise learned beliefs on a toy dataset and evaluate our methods for learning shared representations and structured output prediction, showing trade-offs of learning separate encoders for each information source. Furthermore, we demonstrate how conflict detection and redundancy can increase robustness of inference in a multi-source setting.

System identification of complex and nonlinear systems is a central problem for model predictive control and model-based reinforcement learning. Despite their complexity, such systems can often be approximated well by a set of linear dynamical systems if broken into appropriate subsequences. This mechanism not only helps us find good approximations of dynamics, but also gives us deeper insight into the underlying system. Leveraging Bayesian inference, Variational Autoencoders and Concrete relaxations, we show how to learn a richer and more meaningful state space, e.g. encoding joint constraints and collisions with walls in a maze, from partial and high-dimensional observations. This representation translates into a gain of accuracy of learned dynamics showcased on various simulated tasks.

We propose to learn a hierarchical prior in the context of variational autoencoders to avoid the over-regularisation resulting from a standard normal prior distribution. To incentivise an informative latent representation of the data, we formulate the learning problem as a constrained optimisation problem by extending the Taming VAEs framework to two-level hierarchical models. We introduce a graph-based interpolation method, which shows that the topology of the learned latent representation corresponds to the topology of the data manifold---and present several examples, where desired properties of latent representation such as smoothness and simple explanatory factors are learned by the prior.

Neural samplers such as variational autoencoders (VAEs) or generative adversarial networks (GANs) approximate distributions by transforming samples from a simple random source—the latent space—to samples from a more complex distribution represented by a dataset. While the manifold hypothesis implies that a dataset contains large regions of low density, the training criterions of VAEs and GANs will make the latent space densely covered. Consequently points that are separated by low-density regions in observation space will be pushed together in latent space, making stationary distances poor proxies for similarity. We transfer ideas from Riemannian geometry to this setting, letting the distance between two points be the shortest path on a Riemannian manifold induced by the transformation. The method yields a principled distance measure, provides a tool for visual inspection of deep generative models, and an alternative to linear interpolation in latent space. In addition, it can be applied for robot movement generalization using previously learned skills. The method is evaluated on a synthetic dataset with known ground truth; on a simulated robot arm dataset; on human motion capture data; and on a generative model of handwritten digits.

Robots can rapidly acquire new skills from demonstrations. However, during generalisation of skills or transitioning across fundamentally different skills, it is unclear whether the robot has the necessary knowledge to perform the task. Failing to detect missing information often leads to abrupt movements or to collisions with the environment. Active learning can quantify the uncertainty of performing the task and, in general, locate regions of missing information. We introduce a novel algorithm for active learning and demonstrate its utility for generating smooth trajectories. Our approach is based on deep generative models and metric learning in latent spaces. It relies on the Jacobian of the likelihood to detect non-smooth transitions in the latent space, i.e., transitions that lead to abrupt changes in the movement of the robot. When non-smooth transitions are detected, our algorithm asks for an additional demonstration from that specific region. The newly acquired knowledge modifies the data manifold and allows for learning a latent representation for generating smooth movements. We demonstrate the efficacy of our approach on generalising elementary skills, transitioning across different skills, and implicitly avoiding collisions with the environment. For our experiments, we use a simulated pendulum where we observe its motion from images and a 7-DoF anthropomorphic arm.

Tool-held hitting tasks, like hammering a nail or striking a ball with a bat, require humans, and robots, to purposely collide and transfer momentum from their limbs to the environment. Due to the vibrational dynamics, every tool has a location where a hit is most efficient results in minimal tool vibrations, and consequently maximum energy transfer to the environment. In sports, this location is often referred to as the “sweet spot” of a bat, or racquet. Our recent neuroscience study suggests that humans optimize hits by using the jerk and torque felt at their hand. Motivated by this result, in this work we first analyze the vibrational dynamics of an end-effector-held bat to understand the signature projected by a sweet spot on the jerk and torque sensed at the end-effector. We then use this analysis to develop a controller for a robotic “baseball hitter”. The controller enables the robot-hitter to iteratively adjust its swing trajectory to ensure that the contact with the ball occurs at the sweet spot of the bat. We tested the controller on the DLR LWR III manipulator with three different bats. Like a human, our robot hitter is able to optimize the energy transfer, specifically maximize the ball velocity, during hits, by using its end effector position and torque sensors, and without any prior knowledge of the shape, size or material of the held bat.

The recognition of actions from video sequences has many applications in health monitoring, assisted living, surveillance, and smart homes. Despite advances in sensing, in particular related to 3D video, the methodologies to process the data are still subject to research. We demonstrate superior results by a system which combines recurrent neural networks with convolutional neural networks in a voting approach. The gated-recurrent-unit-based neural networks are particularly well-suited to distinguish actions based on long-term information from optical tracking data; the 3D-CNNs focus more on detailed, recent information from video data. The resulting features are merged in an SVM which then classifies the movement. In this architecture, our method improves recognition rates of state-of-the-art methods by 14% on standard data sets.

The allocation of physical resources to virtual networks, i.e., the virtual network embedding (VNE), is still an on-going research field due to its problem complexity. While many solutions for the online VNE problem exist, only few have focused on methods that can be generally applied for optimization of online embeddings. In this paper, we propose an admission control based on a Recurrent Neural Network (RNN) to improve the overall system performance for the online VNE problem. Before running a VNE algorithm to embed a virtual network request, the RNN predicts whether the request will be accepted by the VNE algorithm based on the current state of the substrate and the virtual network request (VNR). The RNN prevents VNE algorithms from spending time on VNRs that are either infeasible or that cannot be embedded in acceptable time. In order to train and operate the RNN efficiently, we additionally propose new representations for substrate networks and virtual network requests. The representations are based on topological and network resource features to represent the substrate network and the VNRs with low computational complexity. Via simulations, we show that our admission control reduces the overall computational time for the online VNE problem by up to 91 % while preserving VNE performance on average. Using our new substrate and request representations, the RNN achieves an accuracy ranging between 89 % and 98 % for different VNE algorithms, substrate sizes, and VNR arrival rates.

For many tasks, tactile or visual feedback is helpful or even crucial. However, designing controllers that take such high-dimensional feedback into account is non-trivial. Therefore, robots should be able to learn tactile skills through trial and error by using reinforcement learning algorithms. The input domain for such tasks, however, might include strongly correlated or non-relevant dimensions, making it hard to specify a suitable metric on such domains. Auto-encoders specialize in finding compact representations, where defining such a metric is likely to be easier. Therefore, we propose a reinforcement learning algorithm that can learn non-linear policies in continuous state spaces, which leverages representations learned using auto-encoders. We first evaluate this method on a simulated toy-task with visual input. Then, we validate our approach on a real-robot tactile stabilization task.

Convolutional neural networks (CNNs) have recently been very successful in a variety of computer vision tasks, especially on those linked to recognition. Optical flow estimation has not been among the tasks CNNs succeeded at. In this paper we construct CNNs which are capable of solving the optical flow estimation problem as a supervised learning task. We propose and compare two architectures: a generic architecture and another one including a layer that correlates feature vectors at different image locations. Since existing ground truth data sets are not sufficiently large to train a CNN, we generate a large synthetic Flying Chairs dataset. We show that networks trained on this unrealistic data still generalize very well to existing datasets such as Sintel and KITTI, achieving competitive accuracy at frame rates of 5 to 10 fps.

Robust fingertip force detection from fingernail image is a critical strategy that can be applied in many areas. However, prior research fixed many variables that influence the finger color change. This paper analyzes the effect of the finger joint on the force detection in order to deal with the constrained finger position setting. A force estimator method is designed: a model to predict the fingertip force from finger joints measured from 2D cameras and 3 rectangular markers in cooperation with the fingernail images are trained. Then the error caused by the color changes of the joint bending can be avoided. This strategy is a significant step forward from a finger force estimator that requires tedious finger joint setting. The approach is evaluated experimentally. The result shows that it increases the accuracy over 10% for the force in conditions of the finger joint free movement. The estimator is used to demonstrate lifting and replacing objects with various weights.

The most accurate and dependable approach to the in-vivo identification of human limb stiffness is by position perturbation. Moving the limb over a small distance and measuring the effective force gives, when states are steady, direct information about said stiffness. However, existing manipulandi are comparatively slow and/or not very stiff, such that a lumped stiffness is measured. This lumped stiffness includes the limb response during or after reflexes influenced by both, the passive musculotendon and active neuronal component. As this approach usually leads to inconsistencies between the data and the stiffness model, we argue in favour of fast, pre-reflex impedance measurements-i.e., completing the perturbation movement and collecting the data before effects of spinal reflexes or even from the motor cortex can influence the measurements. To obtain such fast planar movements, we constructed a dedicated orthoglide robot while focusing on a lightweight and stiff design. Our subject study of a force task with this device lead to very clean data with always positive definite Cartesian stiffness matrices. By representing them as ellipses, we found them to be substantially bigger in comparison to standard literature which we address to a larger number of recruited motor units. While ellipses orientation and the length of their main axis increased, the shape decreased with the exerted force. The device will be used to derive design criteria for variable-stiffness robots, and to investigate the relation between muscular activity and areflexive joint stiffness for teleoperational approaches.

Estimating human fingertip forces is required to understand force distribution in grasping and manipulation. Human grasping behavior can then be used to develop force-and impedance-based grasping and manipulation strategies for robotic hands. However, estimating human grip force naturally is only possible with instrumented objects or unnatural gloves, thus greatly limiting the type of objects used. In this paper we describe an approach which uses images of the human fingertip to reconstruct grip force and torque at the finger. Our approach does not use finger-mounted equipment, but instead a steady camera observing the fingers of the hand from a distance. This allows for finger force estimation without any physical interference with the hand or object itself, and is therefore universally applicable. We construct a 3-dimensional finger model from 2D images. Convolutional Neural Networks (CNN) are used to predict the 2D image to a 3D model transformation matrix. Two methods of CNN are designed for separate and combined outputs of orientation and position. After learning, our system shows an alignment accuracy over 98% on unknown data. In the final step, a Gaussian process estimates finger force and torque from the aligned images based on color changes and deformations of the nail and its surrounding skin. Experimental results shows that the accuracy achieves about 95% in the force estimation and 90% in the torque.

We introduce a six-actuator robotic joint mechanism with biarticular coupling inspired by the human limb which neither requires pneumatic artificial muscles nor tendon coupling. The actuator can independently change monoarticular and biarticular stiffness as well as both joint positions. We model and analyse the actuator with respect to stiffness variability in comparison with an actuator without biarticular coupling. We demonstrate that the biarticular coupling considerably extends the range of stiffness with an 70-fold improvement in versatility, in particular with respect to the end-point Cartesian stiffness shape and orientation. We suggest using Cartesian stiffness isotropy as an optimisation criterion for future under-actuated versions.

Safety is one of the key issues in the use of robots, especially when human-robot interaction is targeted. Although unforeseen environment situations, such as collisions or unexpected user interaction, can be handled with specially tailored control algorithms, hard- or software failures typically lead to situations where too large torques are controlled, which cause an emergency state: hitting an end stop, exceeding a torque, and so on-which often halts the robot when it is too late. No sufficiently fast and reliable methods exist which can early detect faults in the abundance of sensor and controller data. This is especially difficult since, in most cases, no anomaly data are available. In this paper we introduce a new robot anomaly detection system (RADS) which can cope with abundant data in which no or very little anomaly information is present.

We demonstrate a simple approach with which finger force can be measured from nail coloration. By automatically extracting features from nail images of a finger-mounted CCD camera, we can directly relate these images to the force measured by a force-torque sensor. The method automatically corrects orientation and illumination differences. Using Gaussian processes, we can relate prepro-cessed images of the finger nail to measured force and torque of the finger, allowing us to predict the finger force at a level of 95%–98% accuracy at force ranges up to 10N, and torques around 90% accuracy, based on training data gathered in 90s.

The development of new, light robotic systems has opened up a wealth of human-robot interaction applications. In particular, the use of robot manipulators as personal assistant for the disabled is realistic and affordable, but still requires research as to the brain-computer interface. Based on our previous work with tetraplegic individuals, we investigate the use of low-cost yet stable surface Electromyography (sEMG) interfaces for individuals with Spinal Muscular Atrophy (SMA), a disease leading to the death of neuronal cells in the anterior horn of the spinal cord; with sEMG, we can record remaining active muscle fibers. We show the ability of two individuals with SMA to actively control a robot in 3.5D continuously decoded through sEMG after a few minutes of training, allowing them to regain some independence in daily life. Although movement is not nearly as fast as natural, unimpaired movement, reach and grasp success rates are near 100% after 50s of movement.

Anthropomorphic robots that aim to approach human performance agility and efficiency are typically highly redundant not only in their kinematics but also in actuation. Variable-impedance actuators, used to drive many of these devices, are capable of modulating torque and impedance (stiffness and/or damping) simultaneously, continuously, and independently. These actuators are, however, nonlinear and assert numerous constraints, e.g., range, rate, and effort limits on the dynamics. Finding a control strategy that makes use of the intrinsic dynamics and capacity of compliant actuators for such redundant, nonlinear, and constrained systems is nontrivial. In this study, we propose a framework for optimization of torque and impedance profiles in order to maximize task performance, which is tuned to the complex hardware and incorporating real-world actuation constraints. Simulation study and hardware experiments 1) demonstrate the effects of actuation constraints during impedance control, 2) show applicability of the present framework to simultaneous torque and temporal stiffness optimization under constraints that are imposed by real-world actuators, and 3) validate the benefits of the proposed approach under experimental conditions.

Anthropomorphic robots that aim to approach human performance agility and efficiency are typically highly redundant not only in their kinematics but also in actuation. Variable-impedance actuators, used to drive many of these devices, are capable of modulating torque and passive impedance (stiffness and/or damping) simultaneously and independently. Here, we propose a framework for simultaneous optimisation of torque and impedance (stiffness) profiles in order to optimise task performance, tuned to the complex hardware and incorporating real-world constraints. Simulation and hardware experiments validate the viability of this approach to complex, state dependent constraints and demonstrate task performance benefits of optimal temporal impedance modulation.

In this paper we present a novel and simple handheld device for measuring in vivo human grasp impedance. The measurement method is based on a static identification method and intrinsic impedance is identified inbetween 25 ms. Using this device it is possbile to develop continuous grasp impedance measurement methods as it is an active research topic in physiology as well as in robotics, especially since nowadays (bio-inspired) robotics can be impedance-controlled. Potential applications of human impedance estimation range from impedance-controlled telesurgery to limb prosthetics and rehabilitation robotics. We validate the device through a physiological experiment in which the device is used to show a linear relationship between finger stiffness and grip force.

In this paper we describe and practically demonstrate a robotic arm/hand system that is controlled in real-time in 6D Cartesian space through measured human muscular activity. The soft-robotics control architecture of the robotic system ensures safe physical human robot interaction as well as stable behaviour while operating in an unstructured environment. Muscular control is realised via surface electromyography, a non-invasive and simple way to gather human muscular activity from the skin. A standard supervised machine learning system is used to create a map from muscle activity to hand position, orientation and grasping force which then can be evaluated in real time - the existence of such a map is guaranteed by gravity compensation and low-speed movement. No kinematic or dynamic model of the human arm is necessary, which makes the system quickly adaptable to anyone. Numerical validation shows that the system achieves good movement precision. Live evaluation and demonstration of the system during a robotic trade fair is reported and confirms the validity of the approach, which has potential applications in muscle-disorder rehabilitation or in teleoperation where a close-range, safe master/slave interaction is required, and/or when optical/magnetic position tracking cannot be enforced.

Abstract—One of the main challenges in service robotics is to equip dexterous robotic hands with sensitive tactile sensors in order to cope with the inherent problems posed by unknown and unstructured environments. As the increasing mechatronic integration of complex robotic hands leaves little additional space for proprioceptive sensors, exteroceptive tactile sensors become more and more important. We present a novel tactile sensor design, based on piezo-resistive soft material and a crossed-wire approach. We present the development of a first prototype and its evaluation in various classification tasks, showing promising results. I.

The dexterity of active hand prosthetics is limited not only due to the limited availability of dexterous prosthetic hands, but mainly due to limitations in interfaces. How is an amputee supposed to command the prosthesis what to do (i.e., how to grasp an object) and with what force (i.e., holding a hammer or grasping an egg)? So far, in literature, the most interesting results have been achieved by applying machine learning to forearm surface electromyography (EMG) to classify finger movements; but this approach lacks, in general, the possibility of quantitatively determining the force applied during the grasping act. In this paper we address the issue by applying machine learning to the problem of regression from the EMG signal to the force a human subject is applying to a force sensor. A detailed comparative analysis among three different machine learning approaches (Neural Networks, Support Vector Machines and Locally Weighted Projection Regression) reveals that the type of grasp can be reconstructed with an average accuracy of 90%, and the applied force can be predicted with an average error of 10%, corresponding to about 5N over a range of 50N. None of the tested approaches clearly outperforms the others, which seems to indicate that machine learning as a whole is a viable approach.

We introduce a method based on support vector machines which can detect opening and closing actions of the human thumb, index finger, and other fingers recorded via surface EMG only. The method is shown to be robust across sessions and can be used independently of the position of the arm. With these stability criteria, the method is ideally suited for the control of active prosthesis with a high number of active degrees of freedom. The method is successfully demonstrated on a robotic four-finger hand, and can be used to grasp objects

This paper presents a method for learning and generating rhythmic movement patterns based on a simple central oscillator. It can be used to generate cyclic movements for a robot system which has to solve complex tasks. The system is laid out in such a way that multiple motion dimensions, or degrees of freedom of the robot, are represented independent of each other; therefore, an extension to higher-dimensional problems is easily possible. Guiding the robot by holding its end-effector, the user teaches simple movement primitives forming the basis for a more complex task. Each movement primitive is represented in the system using an oscillator combined with a learned nonlinear mapping. These primitives are then optimally combined to a complete solution to the posed problem. Said optimality is obtained using simulated annealing with the A* global search algorithm. Our approach is demonstrated on the problem of wiping a table, but can be used for many typical problems in service and household robotics.

The authors consider the cerebellum in the vertebrate motor control system. Analyzing the delays in this control loop as well as the complexity of the dynamics of the skeletomuscular system, we find that two well-known interpretations of the function of the cerebellum (that of a Smith predictor, or of an inverse model controller) are insufficient to solve the control problem at hand.

We present a setup to control a four-finger anthropomorphic robot hand using a dataglove. To be able to accurately use the dataglove we implemented a nonlinear learning calibration using a novel neural network technique. Experiments show that a resulting positioning error not exceeding 1.8 mm, but typically 0.5 mm, per finger can be obtained; this accuracy is sufficiently precise for grasping tasks. Based on the dataglove calibration we present a solution for the mapping of human and artificial hand workspaces that enables an operator to intuitively and easily telemanipulate objects with the artificial hand.

Decades of research into the structure and function of the cerebellum have led to a clear understanding of many of its cells, as well as how learning takes place. Furthermore, there are many theories on what signals the cerebellum operates on, and how it works in concert with other parts of the nervous system. Nevertheless, the application of computational cerebellar models to the control of robot dynamics remains in its infant state. To date, a few applications have been realized, but limited to the control of traditional robot structures which, strictly speaking, do not require adaptive control for the tasks that are performed since their dynamic structures are relatively simple. The currently emerging family of light-weight robots (Hirzinger, G. (1996) In Proceedings of the 2nd International Conference on Advanced Robotics, Intelligent Automation, and Active Systems, Vienna, Austria ) poses a new challenge to robot control: owing to their complex dynamics, traditional methods, depending on a full analysis of the dynamics of the system, are no longer applicable since the joints influence each other's dynamics during movement. Can artificial cerebellar models compete here? In this paper, we present a succinct introduction of the cerebellum, and discuss where it could be applied to tackle problems in robotics. Without conclusively answering the above question, an overview of several applications of cerebellar models to robot control is given.

A neural map algorithm has been employed to control a five-joint pneumatic robot arm and gripper through feedback from two video cameras. The pneumatically driven robot arm (SoftArm) employed in this investigation shares essential mechanical characteristics with skeletal muscle systems. To control the position of the arm, 200 neurons formed a network representing the three-dimensional workspace embedded in a four-dimensional system of coordinates from the two cameras, and learned a three-dimensional set of pressures corresponding to the end effector positions, as well as a set of 3/spl times/4 Jacobian matrices for interpolating between these positions. The gripper orientation was achieved through adaptation of a 1/spl times/4 Jacobian matrix for a fourth joint. Because of the properties of the rubber-tube actuators of the SoftArm, the position as a function of supplied pressure is nonlinear, nonseparable, and exhibits hysteresis. Nevertheless, through the neural network learning algorithm the position could be controlled to an accuracy of about one pixel (/spl sim/3 mm) after 200 learning steps and the orientation could be controlled to two pixels after 800 learning steps. This was achieved through employment of a linear correction algorithm using the Jacobian matrices mentioned above. Applications of repeated corrections in each positioning and grasping step leads to a very robust control algorithm since the Jacobians learned by the network have to satisfy the weak requirement that the Jacobian yields a reduction of the distance between gripper and target. The neural network employed in the control of the SoftArm bears close analogies to a network which successfully models visual brain maps. It is concluded, therefore, from this fact and from the close analogy between the SoftArm and natural muscle systems that the successful solution of the control problem has implications for biological visuo-motor control.>

A method is presented to learn 3D grasping of objects with unknown dimensions using a monocular eye-in-hand manipulator. From a sequence of images a motion profile is generated to approach the object of unknown size. It is shown that monocular visual information suffices to control the deceleration of the robot manipulator. A strategy for generating learning samples is presented, and simulation results demonstrate the effectiveness of the method. I. Introduction Sensor based robot control systems can overcome many of the difficulties which are caused by unknown or uncertain models of the environment. Also, conventional sensor based control systems require explicit knowledge of the kinematics and dynamics of the robot arm and a careful calibration of the sensor system. We are interested in self-learning and adaptive systems, where an implicit model of the arm and sensor system is learned from the behaviour of the robot. Neurocomputational techniques have been successfully applied in t...