• 2020-2024close
• EU close
• ZENODO close
• Neuroinformatics close
• Assessing the socio-economic impact... close

Collaborative robots, or cobots, are designed to work alongside humans and to alleviate their physical burdens, such as lifting heavy objects or performing tedious tasks. Ensuring the safety of human–robot interaction (HRI) is paramount for effective collaboration. To achieve this, it is essential to have a reliable dynamic model of the cobot that enables the implementation of torque control strategies. These strategies aim to achieve accurate motion while minimizing the amount of torque exerted by the robot. However, modeling the complex non-linear dynamics of cobots with elastic actuators poses a challenge for traditional analytical modeling techniques. Instead, cobot dynamic modeling needs to be learned through data-driven approaches, rather than analytical equation-driven modeling. In this study, we propose and evaluate three machine learning (ML) approaches based on bidirectional recurrent neural networks (BRNNs) for learning the inverse dynamic model of a cobot equipped with elastic actuators. We also provide our ML approaches with a representative training dataset of the cobot's joint positions, velocities, and corresponding torque values. The first ML approach uses a non-parametric configuration, while the other two implement semi-parametric configurations. All three ML approaches outperform the rigid-bodied dynamic model provided by the cobot's manufacturer in terms of torque precision while maintaining their generalization capabilities and real-time operation due to the optimized sample dataset size and network dimensions. Despite the similarity in torque estimation of these three configurations, the non-parametric configuration was specifically designed for worst-case scenarios where the robot dynamics are completely unknown. Finally, we validate the applicability of our ML approaches by integrating the worst-case non-parametric configuration as a controller within a feedforward loop. We verify the accuracy of the learned inverse dynamic model by comparing it to the actual cobot performance. Our non-parametric architecture outperforms the robot's default factory position controller in terms of accuracy.

AbstractBackgroundWalking impairments are a common consequence of neurological disorders and are assessed with clinical scores that suffer from several limitations. Robot-assisted locomotor training is becoming an established clinical practice. Besides training, these devices could be used for assessing walking ability in a controlled environment. Here, we propose an adaptive assist-as-needed (AAN) control for a treadmill-based robotic exoskeleton, the Lokomat, that reduces the support of the device (body weight support and impedance of the robotic joints) based on the ability of the patient to follow a gait pattern displayed on screen. We hypothesize that the converged values of robotic support provide valid and reliable information about individuals’ walking ability.MethodsFifteen participants with spinal cord injury and twelve controls used the AAN software in the Lokomat twice within a week and were assessed using clinical scores (10MWT, TUG). We used a regression method to identify the robotic measure that could provide the most relevant information about walking ability and determined the test–retest reliability. We also checked whether this result could be extrapolated to non-ambulatory and to unimpaired subjects.ResultsThe AAN controller could be used in patients with different injury severity levels. A linear model based on one variable (robotic knee stiffness at terminal swing) could explain 74% of the variance in the 10MWT and 61% in the TUG in ambulatory patients and showed good relative reliability but poor absolute reliability. Adding the variable ‘maximum hip flexor torque’ to the model increased the explained variance above 85%. This did not extend to non-ambulatory nor to able-bodied individuals, where variables related to stance phase and to push-off phase seem more relevant.ConclusionsThe novel AAN software for the Lokomat can be used to quantify the support required by a patient while performing robotic gait training. The adaptive software might enable more challenging training conditions tuned to the ability of the individuals. While the current implementation is not ready for assessment in clinical practice, we could demonstrate that this approach is safe, and it could be integrated as assist-as-needed training, rather than as assessment.Trial registrationClinicalTrials.gov Identifier: NCT02425332.

This article presents an electromyography-driven musculoskeletal model that can estimate joint torque and joint stiffness simultaneously. We show a novel model parameter calibration procedure that tries to fit reference joint torque and joint stiffness profiles.

The Ethics and Society Subproject has developed this Opinion in order to clarify lessons the Human Brain Project (HBP) can draw from the current discussion of artificial intelligence, in particular the social and ethical aspects of AI, and outline areas where it could usefully contribute. The EU and numerous other bodies are promoting and implementing a wide range of policies aimed to ensure that AI is beneficial - that it serves society. The HBP as a leading project bringing together neuroscience and ICT is in an excellent position to contribute to and to benefit from these discussions. This Opinion therefore highlights some key aspects of the discussion, shows its relevance to the HBP and develops a list of six recommendations.

AbstractDuring interpersonal interactions, people perform actions with different forms of vitality, communicating their positive or negative attitude toward others. For example, a handshake can be “soft” or “vigorous”, a caress can be ‘kind’ or ‘rushed’. While previous studies have shown that the dorso-central insula is a key area for the processing of human vitality forms, there is no information on the perception of vitality forms generated by a humanoid robot. In this study, two fMRI experiments were conducted in order to investigate whether and how the observation of actions generated by a humanoid robot (iCub) with low and fast velocities (Study 1) or replicating gentle and rude human forms (Study 2) may convey vitality forms eliciting the activation of the dorso-central insula. These studies showed that the observation of robotic actions, generated with low and high velocities, resulted in activation of the parieto-frontal circuit typically involved in the recognition and the execution of human actions but not of the insula (Study 1). Most interestingly, the observation of robotic actions, generated by replicating gentle and rude human vitality forms, produced a BOLD signal increase in the dorso-central insula (Study 2). In conclusion, these data highlight the selective role of dorso-central insula in the processing of vitality forms opening future perspectives on the perception and understanding of actions performed by humanoid robots.

The Ethics and Society Subproject has developed this Opinion in order to clarify lessons the Human Brain Project (HBP) can draw from the current discussion of artificial intelligence, in particular the social and ethical aspects of AI, and outline areas where it could usefully contribute. The EU and numerous other bodies are promoting and implementing a wide range of policies aimed to ensure that AI is beneficial - that it serves society. The HBP as a leading project bringing together neuroscience and ICT is in an excellent position to contribute to and to benefit from these discussions. This Opinion therefore highlights some key aspects of the discussion, shows its relevance to the HBP and develops a list of six recommendations. This project has received funding from the European Union's Horizon 2020 Research and Innovation Programme Under Grant Agreement no. 785907