Sense of agency is the subjective feeling of being in control over one’s own actions and their sensory outcomes. Here, we propose that a key feature of sense of agency is its link to intrinsic motivation, which is a key driver for engagement with the environment without external reinforcement. We argue that intrinsic motivation is highest when the sense of agency is experienced at an intermediate level between full control and no control at all. In consequence, we propose that human-operated, artificial intelligence-enabled robotics needs to be designed such that the balance between human sense of agency and the autonomy of the assistive system is achieved, for optimizing the users’ motivation to use the technology, and for improvement of performance. We describe various examples of assistive technologies, ranging from robots in work environments to healthcare robotics, and we discuss how sense of agency is a crucial factor that needs to be accounted for in the engineering endeavor of developing such technologies.

To make best use of large-language models (LLMs) in social robotics it is critical that verbal interaction capabilities are suitably integrated with robot perceptual and behavioral systems, including non-verbal and emotional signaling, such that the robot can use language in a grounded and context-appropriate way. This demonstration shows the integration of a LLM with the layered control architecture of the animal-like robot platform Miro-e alongside deep network models for perception and spoken language recognition and generation. This system is currently being developed as a prototype companion robot for research on robot-assisted therapy.

A better understanding of the nature of human relationships can aid the design of effective and appropriate social behaviour for robots. The investigation of human bonding via robotic modelling can also serve to test psychological theories in an embodied setting. In this work we present a robotic model of "attachment"-the primary bond between child and caregiver that shapes relationship behaviour throughout our lives. Following a dynamical systems approach, we model attachment as a behavioural coupling between motivational oscillators and show, by means of a dynamical analysis, that coupled robot dyads generate dynamical patterns that resemble caregiver-child interactions. By demonstrating coupling in an embodied model, we also show that measures of physical and emotional distance (a psychological variable), inferred from sensory data, can serve as effective control parameters for attachment behaviour. We find that this oscillator framework generates rich patterns of robot behaviours that can be associated with quantitative and qualitative observations of the "strange situation" procedure, an experimental paradigm that is widely studied in human relationship science, and of human avoidant and ambivalent attachment styles. The ability to estimate human attachment style and to generate appropriately-matched robot behaviours could be useful in social and companion robotics.

Robotics can play a useful role in the scientific understanding of the sense of self, both through the construction of embodied models of the self and through the use of robots as experimental probes to explore the human self. In both cases, the embodiment of the robot allows us to devise and test hypotheses about the nature of the self, with regard to its development, its manifestation in behavior, and the diversity of selves in humans, animals, and, potentially, machines. This paper reviews robotics research that addresses the topic of the self—the minimal self, the extended self, and disorders of the self—and highlights future directions and open challenges in understanding the self through constructing its components in artificial systems. An emerging view is that key phenomena of the self can be generated in robots with suitably configured sensor and actuator systems and a layered cognitive architecture involving networks of predictive models.

The rising successes of RL are propelled by combining smart algorithmic strategies and deep architectures to optimize the distribution of returns and visitations over the state-action space. A quantitative framework to compare the learning processes of these eclectic RL algorithms is currently absent but desired in practice. We address this gap by representing the learning process of an RL algorithm as a sequence of policies generated during training, and then studying the policy trajectory induced in the manifold of state-action occupancy measures. Using an optimal transport-based metric, we measure the length of the paths induced by the policy sequence yielded by an RL algorithm between an initial policy and a final optimal policy. Hence, we first define the 'Effort of Sequential Learning' (ESL). ESL quantifies the relative distance that an RL algorithm travels compared to the shortest path from the initial to the optimal policy. Further, we connect the dynamics of policies in the occupancy measure space and regret (another metric to understand the suboptimality of an RL algorithm), by defining the 'Optimal Movement Ratio' (OMR). OMR assesses the fraction of movements in the occupancy measure space that effectively reduce an analogue of regret. Finally, we derive approximation guarantees to estimate ESL and OMR with finite number of samples and without access to an optimal policy. Through empirical analyses across various environments and algorithms, we demonstrate that ESL and OMR provide insights into the exploration processes of RL algorithms and hardness of different tasks in discrete and continuous MDPs.

Robotics is increasingly seen as a useful test bed for computational models of the brain functional architecture underlying animal behavior. We provide an overview of past and current work, focusing on probabilistic and dynamical models, including approaches premised on the free energy principle, situating this endeavor in relation to evidence that the brain constitutes a layered control system. We argue that future neurorobotic models should integrate multiple neurobiological constraints and be hybrid in nature.

We present a biomimetic model of motivated behaviour based on the network architecture of the mammalian hypothalamus and its interaction with brain systems involved in reward, memory and decision-making. Specifically, a novel model of the hypothalamus, viewed as a layered structure, is integrated with a previously-developed model of the hippocampal-striatal network controlling a simulated robot in a navigation task. Hypothalamic modulation of model dopamine signals allows the robot to learn the location of reward while regulating simulated food intake. When ‘satiated’ the robot explores, when ‘hungry’ it moves towards the learned food source. We discuss the potential uses and future challenges of such models in the development of autonomous robots.

It has been previously demonstrated in robots that the mimicking of functional characteristics of biologic memory can be beneficial for providing accurate learning and recognition in circumstances of social human-robot-interaction. The effective encoding of social and physical salient features has been demonstrated through the use of Bayesian Latent Variable Models as abstractions of memories (Simple Synthetic Memories). In this work, we explore the capabilities of formation and recall of tactile memories associated to the encoding of geometric and spatial qualities. Compression and pattern separation are evaluated against the use of raw data in a nearest neighbour regression model, obtaining a substantial improvement in accuracy for prediction of geometric properties of the stimulus. Additionally, pattern completion is assessed with the generation of ‘imagined touch’ streams of data showing similarities to real world tactile observations. The use of this model for tactile memories offers the potential for robustly perform sensorimotor tasks in which the sense of touch is involved.

In developing Trustworthy Autonomous Systems (TAS), as in other domains of technology innovation and research, there is a need to make research processes and activities more accessible to external partners and to the wider public. In this article, we describe the rationale, background and potential for an Open Laboratories approach that complements current strategies in Responsible Research and Innovation and in Open Science, relating this to experience-based aspects of trust in new technologies. We also reflect on the value and benefits of robotic telepresence as an engagement tool that can provide direct access and first-person experience of research, in a manner that is scalable and safe while mitigating some environmental and health concerns.

Introduction Socially assistive robots are devices designed to aid users through social interaction and companionship. Social robotics promise to support cognitive health and aging in place for older adults with and without dementia, as well as their care partners. However, while new and more advanced social robots are entering the commercial market, there are still major barriers to their adoption, including a lack of emotional alignment between users and their robots. Affect Control Theory (ACT) is a framework that allows for the computational modeling of emotional alignment between two partners. Methods We conducted a Canadian online survey capturing attitudes, emotions, and perspectives surrounding pet-like robots among older adults (n = 171), care partners (n = 28), and persons living with dementia (n = 7). Results We demonstrate the potential of ACT to model the emotional relationship between older adult users and three exemplar robots. We also capture a rich description of participants’ robot attitudes through the lens of the Technology Acceptance Model, as well as the most important ethical concerns around social robot use. Conclusions Findings from this work will support the development of emotionally aligned, user-centered robots for older adults, care partners, and people living with dementia.

Seven patients with Welander distal myopathy were subjected to magnetic resonance imaging (MRI) of the lower extremity, and muscle biopsies of the tibialis anterior, soleus and vastus lateralis muscles. MRI revealed abnormalities in both the anterior and posterior compartments of the lower leg in three of the patients, and in only the posterior compartment in the rest of the patients. No MRI abnormalities were found in either the proximal muscles of the leg or in the peroneal or posterior tibial muscle groups. Affected muscles had T1- and T2-values indicating a replacement of muscle fibres with fat tissue. Muscle biopsies showed pathological changes varying from slight to severe in tibialis anterior and soleus muscles in all patients. No muscle fibre abnormalities were seen in the vastus lateralis muscle in any of the patients. In accordance with earlier reports from patients with Welander distal myopathy, there was muscle degeneration of tibialis anterior muscles corresponding to the weakness of dorsal extension of the feet, but also degeneration in the muscles of the posterior compartment. The patients did not, however, show any clinical signs of weakness related to posterior muscle groups. There is no evidence of involvement of proximal muscles of the leg clinically, with MRI or in muscle biopsies.

Hippocampal reverse replay, a phenomenon in which recently active hippocampal cells reactivate in the reverse order, is thought to contribute to learning, particularly reinforcement learning (RL), in animals. Here, we present a novel computational model which exploits reverse replay to improve stability and performance on a homing task. The model takes inspiration from the hippocampal-striatal network, and learning occurs via a three-factor RL rule. To augment this model with hippocampal reverse replay, we derived a policy gradient learning rule that associates place-cell activity with responses in cells representing actions and a supervised learning rule of the same form, interpreting the replay activity as a ‘target’ frequency. We evaluated the model using a simulated robot spatial navigation task inspired by the Morris water maze. Results suggest that reverse replay can improve performance stability over multiple trials. Our model exploits reverse reply as an additional source for propagating information about desirable synaptic changes, reducing the requirements for long-time scales in eligibility traces combined with low learning rates. We conclude that reverse replay can positively contribute to RL, although less stable learning is possible in its absence. Analogously, we postulate that reverse replay may enhance RL in the mammalian hippocampal-striatal system rather than provide its core mechanism.

Active touch sensing can benefit from the representation of uncertainty in order to guide sensing movements and to drive sensing strategies that operate to reduce uncertainty with respect to the task at hand. Here we explore learning approaches that can acquire task knowledge quickly and with relatively small datasets and with the potential to be exploited for active sensing in robots and as models of biological sensory systems. Specifically, we explore the utility of deep (hierarchical) Gaussian Process models (Deep GPs) that have shown promise as models of episodic memory processes due to their low-dimensionality (compactness), generative capability, and ability to explicitly represent uncertainty. Using data obtained in a robotic active touch task (contour following), we show that both single-layer and Deep GP models are capable of providing robust function approximations from tactile data to angle and sensor position, with Deep GPs showing some advantages in terms of accuracy and uncertainty quantification in angle discrimination.

BACKGROUND:Persons living with dementia and their care partners place a high value on aging in place and maintaining independence. Socially assistive robots - embodied characters or pets that provide companionship and aid through social interaction - are a promising tool to support these goals. There is a growing commercial market for these devices, with functions including medication reminders, conversation, pet-like behaviours, and even the collection of health data. While potential users generally report positive feelings towards social robots, persons with dementia have been under-included in design and development, leading to a disconnect between robot functions and the real-world needs and desires of end-users. Furthermore, a key element of social and emotional connectedness in human relationships is emotional alignment - a state where all partners have congruent emotional understandings of a situation. Strong emotional alignment between users and robots will be necessary for social robots to provide meaningful companionship, but a computational model of how to achieve this has been absent from the field. To this end, we propose and test Affect Control Theory (ACT) as a framework to improve emotional alignment between older adults and social robotics.METHOD:Using a Canadian online survey, we introduced respondents to three exemplar social robots with older adult-specific functionalities and evaluated their responses around features, emotions, and ethics using standardized and novel measures (n=171 older adults, n=28 care partners, and n=7 persons living with dementia).RESULT:Overall, participants responded positively to the robots. High priority uses included companionship, interaction, and safety. Reasoning around robot use was pragmatic; curiosity and entertainment were motivators to use, while a perceived lack of need and the mechanical appearance of the robots were detractors. Realistic, cute, and cuddly robots were preferred while artificial-looking, creepy, and toy-like robots were disliked. Most importantly, our evidence supported ACT as a viable model of human-robot emotional alignment.CONCLUSION:This work supports the development of emotionally sophisticated, evidence-based, and user-centered social robotics with older adult- and dementia-specific functionality.

Download the c++ script and build it from the command line using e.g., ‘g++ -o model WilsonPrescott2021.cpp’. Then run the model using e.g., ‘./model log.txt’ to save a summary of the results into the file log.txt for inspection.

This paper proposes a novel unsupervised metric learning approach to detect anomalous/novel objects. Existing object detection approaches either cannot detect novel categories or require human annotations even in a small scale (such as few-shot learning). To overcome this, especially for robotic applications where human annotations are not available, this project leverages unsupervised representation learning and unsupervised metric learning to discover feature prototypes of unknown fine-grained categories i.e. clusters in the low-dimensional embedding space. Specifically, the proposed approach leverages deep clustering and self-reconstruction to learn feature prototypes for normal objects. More importantly, we interpolate the latent features and generate pseudo anomalous examples to learn the embedding space of good compactness and sparseness to learn a discriminative embedding space, facilitating distinguishing anomalous examples from normal ones. The learned prototypes can be further used to infer the probability of novel objects using the metric distance to the prototypes. The proposed unsupervised learning approach is also integrated with a Region Proposal Network as a detection pipeline and real-time detection is achieved. This paper uses the StreetHazards dataset of CAOS benchmark for training and evaluation and comparison experiments are implemented to demonstrate the effectiveness of the proposed approach.

Title: Preparing the Workforce for 2030: Skills and Education for Robotics & Autonomous Systems Year: 2021 Citation: (2021). Preparing the Workforce for 2030: Skills and Education for Robotics & Autonomous Systems. EPSRC UK-RAS Network. doi: 10.31256/WP2021.1. ISSN: 2398-4422 (Online), ISSN 2398-4414 (Print) Download PDF