Search

Soft robots are becoming extremely popular thanks to their intrinsic safety to contacts and adaptability. However, the potentially infinite number of Degrees of Freedom makes their modeling a daunting task, and in many cases only an approximated description is available. This challenge makes reinforcement learning (RL) based approaches inefficient when deployed on a realistic scenario, due to the large domain gap between models and the real platform. In this work, we demonstrate, for the first time, how Domain Randomization (DR) can solve this problem by enhancing RL policies with: i) a higher robustness w.r.t. environmental changes; ii) a higher affordability of learned policies when the target model differs significantly from the training model; iii) a higher effectiveness of the policy, which can even autonomously learn to exploit the environment to increase the robot capabilities (environmental constraints exploitation). Moreover, we introduce a novel algorithmic extension of previous adaptive domain randomization methods for the automatic inference of dynamics parameters for deformable objects. We provide results on four different tasks and two soft robot designs, opening interesting perspectives for future research on Reinforcement Learning for closed-loop soft robot control., Project website at https://andreaprotopapa.github.io/dr-soro/

In the last decade, most research in Machine Learning contributed to the improvement of existing models, with the aim of increasing the performance of neural networks for the solution of a variety of different tasks. However, such advancements often come at the cost of an increase of model memory and computational requirements. This represents a significant limitation for the deployability of research output in realistic settings, where the cost, the energy consumption, and the complexity of the framework play a crucial role. To solve this issue, the designer should search for models that maximise the performance while limiting its footprint. Typical approaches to reach this goal rely either on manual procedures, which cannot guarantee the optimality of the final design, or upon Neural Architecture Search algorithms to automatise the process, at the expenses of extremely high computational time. This paper provides a solution for the fast identification of a neural network that maximises the model accuracy while preserving size and computational constraints typical of tiny devices. Our approach, named FreeREA, is a custom cell-based evolution NAS algorithm that exploits an optimised combination of training-free metrics to rank architectures during the search, thus without need of model training. Our experiments, carried out on the common benchmarks NAS-Bench-101 and NATS-Bench, demonstrate that i) FreeREA is a fast, efficient, and effective search method for models automatic design; ii) it outperforms State of the Art training-based and training-free techniques in all the datasets and benchmarks considered, and iii) it can easily generalise to constrained scenarios, representing a competitive solution for fast Neural Architecture Search in generic constrained applications. The code is available at \url{https://github.com/NiccoloCavagnero/FreeREA}., 16 pages, 4 figures

Physics simulators have shown great promise for conveniently learning reinforcement learning policies in safe, unconstrained environments. However, transferring the acquired knowledge to the real world can be challenging due to the reality gap. To this end, several methods have been recently proposed to automatically tune simulator parameters with posterior distributions given real data, for use with domain randomization at training time. These approaches have been shown to work for various robotic tasks under different settings and assumptions. Nevertheless, existing literature lacks a thorough comparison of existing adaptive domain randomization methods with respect to transfer performance and real-data efficiency. In this work, we present an open benchmark for both offline and online methods (SimOpt, BayRn, DROID, DROPO), to shed light on which are most suitable for each setting and task at hand. We found that online methods are limited by the quality of the currently learned policy for the next iteration, while offline methods may sometimes fail when replaying trajectories in simulation with open-loop commands. The code used will be released at https://github.com/gabrieletiboni/adr-benchmark., 15 pages, 6 figures

Wearable sensing has emerged as a promising solution for enabling unobtrusive and ergonomic measurements of the human motion. However, the reconstruction performance of these devices strongly depends on the quality and the number of sensors, which are typically limited by wearability and economic constraints. A promising approach to minimize the number of sensors is to exploit dimensionality reduction approaches that fuse prior information with insufficient sensing signals, through minimum variance estimation. These methods were successfully used for static hand pose reconstruction, but their translation to motion reconstruction has not been attempted yet. In this work, we propose the usage of functional principal component analysis to decompose multimodal, time-varying motion profiles in terms of linear combinations of basis functions. Functional decomposition enables the estimation of the a priori covariance matrix, and hence the fusion of scarce and noisy measured data with a priori information. We also consider the problem of identifying which elemental variables to measure as the most informative for a given class of tasks. We applied our method to two different datasets of upper limb motion D1 (joint trajectories) and D2 (joint trajectories + EMG data) considering an optimal set of measures (four joints for D1 out of seven, three joints, and eight EMGs for D2 out of seven and twelve, respectively). We found that our approach enables the reconstruction of upper limb motion with a median error of 0.013 ± 0.006 rad for D1 (relative median error 0.9%), and 0.038 ± 0.023 rad and 0.003 ± 0.002 mV for D2 (relative median error 2.9% and 5.1%, respectively).Wearable sensing has emerged as a promising solution for enabling unobtrusive and ergonomic measurements of the human motion. However, the reconstruction performance of these devices strongly depends on the quality and the number of sensors, which are typically limited by wearability and economic constraints. A promising approach to minimize the number of sensors is to exploit dimensionality reduction approaches that fuse prior information with insufficient sensing signals, through minimum variance estimation. These methods were successfully used for static hand pose reconstruction, but their translation to motion reconstruction has not been attempted yet. In this work, we propose the usage of functional principal component analysis to decompose multimodal, time-varying motion profiles in terms of linear combinations of basis functions. Functional decomposition enables the estimation of the a priori covariance matrix, and hence the fusion of scarce and noisy measured data with a priori information. We also consider the problem of identifying which elemental variables to measure as the most informative for a given class of tasks. We applied our method to two different datasets of upper limb motion D1 (joint trajectories) and D2 (joint trajectories + EMG data) considering an optimal set of measures (four joints for D1 out of seven, three joints, and eight EMGs for D2 out of seven and twelve, respectively). We found that our approach enables the reconstruction of upper limb motion with a median error of 0.013 ± 0.006 rad for D1 (relative median error 0.9%), and 0.038 ± 0.023 rad and 0.003 ± 0.002 mV for D2 (relative median error 2.9% and 5.1%, respectively).

In this report, we describe the technical details of our submission to the EPIC-Kitchens-100 Unsupervised Domain Adaptation (UDA) Challenge in Action Recognition. To tackle the domain-shift which exists under the UDA setting, we first exploited a recent Domain Generalization (DG) technique, called Relative Norm Alignment (RNA). Secondly, we extended this approach to work on unlabelled target data, enabling a simpler adaptation of the model to the target distribution in an unsupervised fashion. To this purpose, we included in our framework UDA algorithms, such as multi-level adversarial alignment and attentive entropy. By analyzing the challenge setting, we notice the presence of a secondary concurrence shift in the data, which is usually called environmental bias. It is caused by the existence of different environments, i.e., kitchens. To deal with these two shifts (environmental and temporal), we extended our system to perform Multi-Source Multi-Target Domain Adaptation. Finally, we employed distinct models in our final proposal to leverage the potential of popular video architectures, and we introduced two more losses for the ensemble adaptation. Our submission (entry 'plnet') is visible on the leaderboard and ranked in 2nd position for 'verb', and in 3rd position for both 'noun' and 'action'., Comment: 3rd place in the 2022 EPIC-KITCHENS-100 Unsupervised Domain Adaptation Challenge for Action Recognition. arXiv admin note: substantial text overlap with arXiv:2107.00337

The need for users’ safety and technology acceptability has incredibly increased with the deployment of co-bots physically interacting with humans in industrial settings, and for people assistance. A well-studied approach to meet these requirements is to ensure human-like robot motions and interactions. In this manuscript, we present a research approach that moves from the understanding of human movements and derives usefull guidelines for the planning of arm movements and the learning of skills for physical interaction of robots with the surrounding environment. &nbsp

Objective. Brain–computer interfaces (BCIs) exploit computational features from brain signals to perform a given task. Despite recent neurophysiology and clinical findings indicating the crucial role of functional interplay between brain and cardiovascular dynamics in locomotion, heartbeat information remains to be included in common BCI systems. In this study, we exploit the multidimensional features of directional and functional interplay between electroencephalographic and heartbeat spectra to classify upper limb movements into three classes. Approach. We gathered data from 26 healthy volunteers that performed 90 movements; the data were processed using a recently proposed framework for brain–heart interplay (BHI) assessment based on synthetic physiological data generation. Extracted BHI features were employed to classify, through sequential forward selection scheme and k-nearest neighbors algorithm, among resting state and three classes of movements according to the kind of interaction with objects. Main results. The results demonstrated that the proposed brain–heart computer interface (BHCI) system could distinguish between rest and movement classes automatically with an average 90% of accuracy. Significance. Further, this study provides neurophysiology insights indicating the crucial role of functional interplay originating at the cortical level onto the heart in the upper limb neural control. The inclusion of functional BHI insights might substantially improve the neuroscientific knowledge about motor control, and this may lead to advanced BHCI systems performances.

Notwithstanding the enormous advancements in the design of mechatronic devices, handling the complexity of modern robots is still an open issue. Having multiple, redundant, degrees of freedom could be a major limitation for autonomous robots because of the large computational burden, while for prosthetic applications reducing (via hardware or software) the number of degrees of actuation is mandatory to enable the control of the device using bio-signals. A promising approach to solve this issue is to take inspiration from hu- man motion control and implement dimensionality reduction techniques. In this extended abstract we, then, discuss recent advancements on dimensionality reduction-based methods for planning of humanoid manipulators.

THE execution of reliable and stable grasping with artificial hands is a main challenge in the robotics field, due to its practical relevance and theoretical complexity. The classic approach used to grasp with rigid robotic hands generally favored object-centric analytical solutions. More specifically, a set of available contact points is hypothesized, while their position and contact forces are evaluated from the object knowledge [1]. Although very elegant and theoretically sound, this approach has not yet produced the desired outcomes in practice. To address these limitations, in soft artificial hands part of the control intelligence has been directly embedded in their mechanism, through the purposeful introduction of elastic elements and under-actuation patterns [2], [3]. Thanks to their intrinsic compliance, soft hands can mold around the external items and exploit their environment, thus multiplying their grasping capabilities, similarly to humans. Several papers have shown that soft end-effectors can achieve high-level grasping performance when operated by humans (see e.g. [4], [5]). However, such level of dexterity is still unmatched in autonomous grasp execution. This represents an important challenge not only for industrial applications but also for advanced human-robot interaction. Indeed, as of today, robots are no more in their cages – where they originally acted in industrial pipelines for accomplishing precise yet repetitive tasks – but actively interact with humans as assistive devices or co-bots, facing action execution in unstructured environments. Under this regard, the compliance of the mechanical structure of soft systems and hands not only enables a safe interaction with humans, but it also allows a purposeful adaptation with respect to different object shapes, for manipulation purposes. However, classic approaches cannot be applied to this kind of hands, which - by their own nature - do not allow fingertips placement with the required precision and relative independence. On the contrary, data driven approaches could be the key to learn from humans how to manage soft hands, towards higher levels of autonomous grasping capabilities., https://youtu.be/tFTw1Q3fhRw

The paper describes the activities of the European project SOPHIA, Socio-Physical Interaction Skills for Cooperative Human-Robot Systems in Agile Production. The consortium involves European partners from academia, research organizations and industry. Themain goal of the project is to develop a newgeneration of CoBots andWearbots and advanced instrumental-based biomechanical risk assessment tools in industrial scenarios to reduce work-related musculoskeletal disorders and to improve productivity in industry 4.0. Further aimof the project is to create the basis for newergonomic international Standards for manual handling activities.

The accurate assessment of upper limb motion impairment induced by stroke - which represents one of the primary causes of disability world-wide - is the first step to successfully monitor and guide patients' recovery. As of today, the majority of the procedures relies on clinical scales, which are mostly based on ordinal scaling, operator-dependent, and subject to floor and ceiling effects. In this work, we intend to overcome these limitations by proposing a novel approach to analytically evaluate the level of pathological movement coupling, based on the quantification of movement complexity. To this goal, we consider the variations of functional Principal Components applied to the reconstruction of joint angle trajectories of the upper limb during daily living task execution, and compared these variations between two conditions, i.e. the affected and non-affected arm. A Dissimilarity Index, which codifies the severity of the upper limb motor impairment with respect to the movement complexity of the non-affected arm, is then proposed. This methodology was validated as a proof of concept upon a set of four chronic stroke subjects with mild to moderate arm and hand impairments. As a first step, we evaluated whether the derived outcomes differentiate between the two conditions upon the whole data-set. Secondly, we exploited this concept to discern between different subjects and impairment levels. Results show that: i) differences in terms of movement variability between the affected and nonaffected upper limb are detectable and ii) different impairment profiles can be characterized for single subjects using the proposed approach. Although provisional, these results are very promising and suggest this approach as a basis ingredient for the definition of a novel, operator-independent, sensitive, intuitive and widely applicable scale for the evaluation of upper limb motion impairment.