Abstracts:This paper describes how to achieve highly accurate unsupervised spatial lexical acquisition from speech-recognition results including phoneme recognition errors. In most research into lexical acquisition, the robot has no pre-existing lexical knowledge. The robot acquires sequences of some phonemes as words from continuous speech signals. In a previous study, we proposed a nonparametric Bayesian spatial concept acquisition method (SpCoA) that integrates the robot’s position and words obtained by unsupervised word segmentation from uncertain syllable recognition results. However, SpCoA has a very critical problem to be solved in lexical acquisition; the boundaries of word segmentation are incorrect in many cases because of many phoneme recognition errors. Therefore, we propose an unsupervised machine learning method (SpCoA++) for the robust lexical acquisition of novel words relating to places visited by the robot. The proposed SpCoA++ method performs an iterative estimation of learning spatial concepts and updating a language model using place information. SpCoA++ can select a candidate including many words that better represent places from multiple word-segmentation results by maximizing the mutual information between segmented words and spatial concepts. The experimental results demonstrate a significant improvement of the phoneme accuracy rate of learned words relating to place in the proposed method by word-segmentation results based on place information, in comparison to the conventional methods. We indicate that the proposed method enables the robot to acquire words from speech signals more accurately, and improves the estimation accuracy of the spatial concepts.

Abstracts:We propose an approach for instructing a robot using natural language to solve complex tasks in a dynamic environment. In this study, we elaborate on a framework that allows a humanoid robot to understand natural language, derive symbolic representations of its sensorimotor experience, generate complex plans according to the current world state, and monitor plan execution. The presented development supports replacing missing objects and suggesting possible object locations. It is a realization of the concept of structural bootstrapping developed in the context of the European project Xperience. The framework is implemented within the robot development environment ArmarX. We evaluate the framework on the humanoid robot ARMAR-III in the context of two experiments: a demonstration of the real execution of a complex task in the kitchen environment on ARMAR-III and an experiment with untrained users in a simulation environment.

Abstracts:Map building of indoor environments is considered a basic building block for autonomous mobile robots, enabling, among others, self-localization and efficient path planning. While the mainstream approach stores maps as occupancy grids of regular cells, some works have advocated for the use of maps composed of line segments to represent the boundary of obstacles, leveraging on their more compact size. In order to limit both the growth of the corresponding data structures and the effort in processing these maps, a number of methods have been proposed for merging together redundant line segments that represent the same portion of the environment. In this paper, we experimentally compare some of the most significant methods for merging line segments in maps by applying them to publicly available data sets. At the end, we propose some guidelines to choose the appropriate method.

Abstracts:This paper presents a position control scheme for flexible joint robot (FJR). Traditional energy shaping controller with gravity compensation is revisited first and some drawbacks are analyzed; on this basis, a nonlinear state feedback controller along with energy shaping is provided, which can enhance the residual vibration suppression and reduce the overshoots of motor position. Boundedness analysis is presented and global convergence is analytically proven. Experiment results illustrate effectiveness of the proposed scheme.

Abstracts:Cooperative manipulation of a rigid object is challenging and represents an interesting and active research area, especially when these robots are subject to joint and task prioritization constraints. In cooperative manipulation, a primary task is to maintain the coordination of motions, to avoid severe damage caused by the violation of kinematic constraints imposed by the closed chain mechanism. This paper proposes a kinematic controller for dual-arm cooperative manipulation that ensures safety by providing relative coordinated motion as highest priority task and joint limit avoidance and world-space trajectory following at a lower priority. The coordination of motions is based on modular relative Jacobian formulation. The approach is applicable to systems composed of redundant or non-redundant manipulators. Experiments in simulation demonstrate the behavior of the approach under different redundancy configurations. Experiments on two robots with different number of redundant motions show the applicability of the proposed approach to cooperative manipulation under joint limit constraints.

Abstracts:This paper presents an efficient approach to contact force distribution, which is aimed at computing optimal contact forces to generate the required wrench for grasping an object. It has been derived in the previous work that this problem can be reduced to computing the intersection of the ray originating from the origin along the required wrench with the boundary of the grasp wrench set. Noticing that the grasp wrench set is fixed once contact positions are determined, we propose an algorithm to pre-compute a hierarchy of polytopes in the grasp wrench set and a list of facets from the interior of the grasp wrench set to its boundary. Then, the ray’s intersection can be quickly found by searching the list of facets and optimal contact forces can be computed in real time. Numerical examples show that the online computation of the proposed approach is one order of magnitude faster than the latest algorithm to compute the ray’s intersection and two orders of magnitude faster than general-purpose optimization algorithms. This approach to contact force distribution is an iterative solution that can run until reaching a desired accuracy.

Abstracts:The concept of postural synergy is recently being widely applied to the mechanical design of prosthesis. However, there are few research literatures on the application of postural synergy in the mechanical design of an exoskeletal rehabilitation robot. In this paper, the reaching movements of human arm are analysed by the principal component analysis method and two most significant synergies of the joints of human arm, which can account for more than 80% of the variation, are extracted. We propose a postural synergy based method to design the kinematic transmission mechanism for a multi-joint upper-limb exoskeletal rehabilitation robot with two actuators. Additionally, the configuration of passive joints and balance weights is put forward to improve the performance of the robot. The postural synergy based design method is formulated for replicating human arm reaching movements. Finally, tests are taken on the prototype to validate the proposed method. Furthermore, the proposed method can be potentially extended to the design of underactuated robotic arm or manipulator which is just required to perform the specific types of movements.

Abstracts:Cyber-security for robotic systems is a growing concern. Many mobile robots rely heavily on Real Time Location Systems to operate safely in different environments. As a result, Real Time Location Systems have become a vector of attack for robots and autonomous systems, a situation which has not been studied well. This article shows that cyber-attacks on Real Time Location Systems can be detected by a system built using supervised learning. Furthermore it shows that some type of cyber-attacks on Real Time Location Systems, specifically Denial of Service and Spoofing, can be detected by a system built using Machine Learning techniques. In order to construct models capable of detecting those attacks, different supervised learning algorithms have been tested and validated using a dataset of real data recorded by a wheeled robot and a commercial Real Time Location System, based on Ultra Wideband beacons. Experimental results with a cross-validation analysis have shown that Multi-Layer Perceptron classifiers get the highest test score and the lowest validation error. Moreover, it is the model with less overfitting and more sensitivity for detecting Denial of Service and Spoofing cyber-attacks on Real Time Location Systems.

Abstracts:This article presents the development of a power augmentation and rehabilitation exoskeleton based on a novel actuator. The proposed soft actuators are extensor bending pneumatic artificial muscles. This type of soft actuator is derived from extending McKibben artificial muscles by reinforcing one side to prevent extension. This research has experimentally assessed the performance of this new actuator and an output force mathematical model for it has been developed. This new mathematical model based on the geometrical parameters of the extensor bending pneumatic artificial muscle determines the output force as a function of the input pressure. This model is examined experimentally for different actuator sizes. After promising initial experimental results, further model enhancements were made to improve the model of the proposed actuator. To demonstrate the new bending actuators a power augmentation and rehabilitation soft glove has been developed. This soft hand exoskeleton is able to fit any adult hand size without the need for any mechanical system changes or calibration. EMG signals from the human hand have been monitored to prove the performance of this new design of soft exoskeleton. This power augmentation and rehabilitation wearable robot has been shown to reduce the amount of muscles effort needed to perform a number of simple grasps.

Abstracts:Although in recent years there have been quite a few studies aimed at the navigation of robots in cluttered environments, few of these have addressed the problem of robots navigating while moving a large or heavy objects. This is especially useful when transporting loads with variable weights and shapes without having to change the robot hardware. Inspired by the wide use of makeshift carts by humans, we tackle, in this work, the problem of a humanoid robot navigating in a cluttered environment while displacing a heavy load that lies on a cart-like object. We present a complete navigation scheme, from the incremental construction of a map of the environment and the computation of collision-free trajectories to the control to execute these trajectories. Our contributions are as follows: (1) a whole-body control scheme that makes the humanoid use its hands and arms to control the motions of the cart–load system (e.g. tight turns) (2) a sensorless approach to automatically select the appropriate primitive set according to the load weight (3) a motion planning algorithm to find an obstacle-free trajectory using the appropriate primitive set and the constructed map of the environment as input (4) an efficient filtering technique to remove the cart from the field of view of the robot while improving the general performances of the SLAM algorithms and (5) a continuous and consistent odometry data formed by fusing the visual and the robot odometry information. We present experiments conducted on a real Nao robot, equipped with an RGB-D sensor mounted on its head, pushing a cart with different loads. Our experiments show that the payload can be significantly increased without changing the robot’s main hardware, and therefore enacting the capacity of humanoid robots in real-life situations.