Aligning large language models (LLMs) behaviour with human intent is critical for future AI. An important yet often overlooked aspect of this alignment is the perceptual alignment. Perceptual modalities like touch are more multifaceted and nuanced compared to other sensory modalities such as vision. This study investigates how well LLMs can understand and interpret human touch experiences by focusing on their capacity to perceive the tactile qualities of everyday objects. For instance, it assesses whether LLMs can recognize that silk satin is softer and smoother than cotton denim. We developed a “Guess What Textile“ interaction using a custom AI system that enables participants to narrate their touch experiences in the “textile hand” task. Participants were given two textile samples–a target and a reference–to handle. Without seeing them, participants described the differences between them to the LLM. Using these descriptions, the LLM attempted to identify the target textile by assessing similarity within its high-dimensional embedding space, where its perceptual representations are encoded. Our results suggest that a degree of perceptual alignment exists; however, it varies significantly among different textile samples. For example, LLM predictions are well aligned for silk satin, but not for cotton denim. Moreover, participants felt that their textile experiences were not closely matched by the LLM predictions. This study is the first exploration into perceptual alignment around touch using LLM encoders, exemplified through textile hand task. We discuss possible sources of this alignment variance, and how better human-AI perceptual alignment can benefit future everyday tasks. • We address the gap in understanding perception alignment between human touch and AI. • First study on alignment between human touch experiences and LLMs in embeddings. • A novel interactive task probes LLMs’ learned representations for human alignment. • LLMs show perceptual biases, aligning better with certain textiles than others.

Decision Transformers (DTs) effectively capture long-range dependencies using self-attention but struggle with fine-grained local relationships, especially the Markovian properties in many offline-RL datasets. Conversely, Decision Convformer (DC) utilizes convolutional filters for capturing local patterns but shows limitations in tasks demanding long-term dependencies, such as Maze2d. To address these limitations and leverage both strengths, we propose the Long-Short Decision Transformer (LSDT), a general-purpose architecture to effectively capture global and local dependencies across two specialized parallel branches (self-attention and convolution). We explore how these branches complement each other by modeling various ranged dependencies across different environments, and compare it against other baselines. Experimental results demonstrate our LSDT achieves state-of-the-art performance and notable gains over the standard DT in D4RL offline RL benchmark. Leveraging the parallel architecture, LSDT performs consistently on diverse datasets, including Markovian and non-Markovian. We also demonstrate the flexibility of LSDT's architecture, where its specialized branches can be replaced or integrated into models like DC to improve their performance in capturing diverse dependencies. Finally, we also highlight the role of goal states in improving decision-making for goal-reaching tasks like Antmaze.

Human-artificial intelligence (AI) alignment ensures that AI systems align with human goals and behaviors. This paper introduces perceptual alignment as a critical aspect of this alignment, focusing on the concurrence between human judgments and AI evaluations across sensory modalities. We particularly explore how Multimodal Large Language Models (MLLMs), which process both visual and textual data, interpret the tactile qualities of textiles—a significant challenge in online shopping environments. Our research analyzes six vision-based MLLMs to see how they describe the tactile experience of textiles and compares these AI-generated descriptions with human assessments. Through semantic similarity measures and in-person evaluations, we investigate the extent of alignment between human perceptions and AI descriptions. Our findings indicate significant variability in the AI’s ability to interpret different textiles, highlighting both the potential and limitations of current AI models in achieving perceptual alignment. This work contributes to understanding the complexities of aligning AI capabilities with human touch sensory experiences.

Virtual Reality (VR) has revolutionized how we interact with digital worlds. However, programming for VR remains a complex and challenging task, requiring specialized skills and knowledge. Powered by large language models (LLMs), DreamCodeVR is designed to assist users, irrespective of their coding skills, in crafting basic object behavior in VR environments by translating spoken language into code within an active application. This approach seeks to simplify the process of defining behaviors visual changes through speech. Our preliminary user study indicated that the system’s speech interface supports elementary programming tasks, highlighting its potential to improve accessibility for users with varying technical skills. However, it also uncovered a wide range of challenges and opportunities. In an extensive discussion, we detail the system’s strengths, weaknesses, and areas for future research.

Duration modelling has become an important research problem once more with the rise of non-attention neural textto-speech systems.The current approaches largely fall back to relying on previous statistical parametric speech synthesis technology for duration prediction, which poorly models the expressiveness and variability in speech.In this paper, we propose two alternate approaches to improve duration modelling.First, we propose a duration model conditioned on phrasing that improves the predicted durations and provides better modelling of pauses.We show that the duration model conditioned on phrasing improves the naturalness of speech over our baseline duration model.Second, we also propose a multi-speaker duration model called Cauliflow, that uses normalising flows to predict durations that better match the complex target duration distribution.Cauliflow performs on par with our other proposed duration model in terms of naturalness, whilst providing variable durations for the same prompt and variable levels of expressiveness.Lastly, we propose to condition Cauliflow on parameters that provide an intuitive control of the pacing and pausing in the synthesised speech in a novel way.

Major virtual reality (VR) companies are trying to enhance the sense of immersion in virtual environments by implementing haptic feedback in their systems (e.g., Oculus Touch). It is known that tactile stimulation adds realism to a virtual environment. In addition, when users are not limited by wearing any attachments (e.g., gloves), it is even possible to create more immersive experiences. Mid-air haptic technology provides contactless haptic feedback and offers the potential for creating such immersive VR experiences. However, one of the limitations of mid-air haptics resides in the need for freehand tracking systems (e.g., Leap Motion) to deliver tactile feedback to the user's hand. These tracking systems are not accurate, limiting designers capability of delivering spatially precise tactile stimulation. Here, we investigated an alternative way to convey incongruent visual-tactile stimulation that can be used to create the illusion of a congruent visual-tactile experience, while participants experience the phenomenon of the rubber hand illusion in VR.

In virtual reality, the lack of kinesthetic feedback often prevents users from experiencing the weight of virtual objects. Control-to-display (C/D) ratio manipulation has been proposed as a method to induce weight perception without kinesthetic feedback. Based on the fact that lighter (heavier) objects are easier (harder) to move, this method induces an illusory perception of weight by manipulating the rendered position of users' hands---increasing or decreasing their displayed movements. In a series of experiments we demonstrate that C/D-ratio induces a genuine perception of weight, while preserving ownership over the virtual hand. This means that such a manipulation can be easily introduced in current VR experiences without disrupting the sense of presence. We discuss these findings in terms of estimation of physical work needed to lift an object. Our findings provide the first quantification of the range of C/D-ratio that can be used to simulate weight in virtual reality.

Virtual Reality allow users to explore virtual environments naturally, by moving their head and body. However, the size of the environments they can explore is limited by real world constraints, such as the tracking technology or the physical space available. Existing techniques removing these limitations often break the metaphor of natural navigation in VR (e.g. steering techniques), involve control commands (e.g., teleporting) or hinder precise navigation (e.g., scaling user's displacements). This paper proposes NaviFields, which quantify the requirements for precise navigation of each point of the environment, allowing natural navigation within relevant areas, while scaling users' displacements when travelling across non-relevant spaces. This expands the size of the navigable space, retains the natural navigation metaphor and still allows for areas with precise control of the virtual head. We present a formal description of our NaviFields technique, which we compared against two alternative solutions (i.e., homogeneous scaling and natural navigation). Our results demonstrate our ability to cover larger spaces, introduce minimal disruption when travelling across bigger distances and improve very significantly the precise control of the viewpoint inside relevant areas.

The use of the senses of vision and audition as interactive means has dominated the field of Human-Computer Interaction (HCI) for decades, even though nature has provided us with many more senses for perceiving and interacting with the world around us. That said, it has become attractive for HCI researchers and designers to harness touch, taste, and smell in interactive tasks and experience design. In this paper, we present research and design insights gained throughout an interdisciplinary collaboration on a six-week multisensory display – Tate Sensorium – exhibited at the Tate Britain art gallery in London, UK. This is a unique and first time case study on how to design art experiences whilst considering all the senses (i.e., vision, sound, touch, smell, and taste), in particular touch, which we exploited by capitalizing on a novel haptic technology, namely, mid-air haptics. We first describe the overall set up of Tate Sensorium and then move on to describing in detail the design process of the mid-air haptic feedback and its integration with sound for the Full Stop painting by John Latham (1961). This was the first time that mid-air haptic technology was used in a museum context over a prolonged period of time and integrated with sound to enhance the experience of visual art. As part of an interdisciplinary team of curators, sensory designers, sound artists, we selected a total of three variations of the mid-air haptic experience (i.e., haptic patterns), which were alternated at dedicated times throughout the six-week exhibition. We collected questionnaire-based feedback from 2500 visitors and conducted 50 interviews to gain quantitative and qualitative insights on visitors’ experiences and emotional reactions. Whilst the questionnaire results are generally very positive with only a small variation of the visitors’ arousal ratings across the three tactile experiences designed for the Full Stop painting, the interview data shed light on the differences in the visitors’ subjective experiences. Our findings suggest multisensory designers and art curators can ensure a balance between surprising experiences versus the possibility of free exploration for visitors. In addition, participants expressed that experiencing art with the combination of mid-air haptic and sound was immersive and provided an up-lifting experience of touching without touch. We are convinced that the insights gained from this large-scale and real-world field exploration of multisensory experience design exploiting a new and emerging technology provide a solid starting point for the HCI community, creative industries, and art curators to think beyond conventional art experiences. Specifically, our work demonstrates how novel mid-air technology can make art more emotionally engaging and stimulating, especially abstract art that is often open to interpretation.

Touch is a powerful vehicle for communication between humans. The way we touch (how) embraces and mediates certain emotions such as anger, joy, fear, or love. While this phenomenon is well explored for human interaction, HCI research is only starting to uncover the fine granularity of sensory stimulation and responses in relation to certain emotions. Within this paper we present the findings from a study exploring the communication of emotions through a haptic system that uses tactile stimulation in mid-air. Here, haptic descriptions for specific emotions (e.g., happy, sad, excited, afraid) were created by one group of users to then be reviewed and validated by two other groups of users. We demonstrate the non-arbitrary mapping between emotions and haptic descriptions across three groups. This points to the huge potential for mediating emotions through mid-air haptics. We discuss specific design implications based on the spatial, directional, and haptic parameters of the created haptic descriptions and illustrate their design potential for HCI based on two design ideas.

Haptic perception constitutes an important component of our everyday interaction with many products. At the same time, several studies have, in recent years, demonstrated the importance of involving the emotions in the user-product interaction process. The present study was designed to investigate whether haptic interactions can affect, or modulate, people's responses to standardized emotional stimuli. 36 participants completed a self-assessment test concerning their emotional state utilizing as a pointer either a PHANToM device simulating a viscous force field while they moved the stylus, or else a stylus with no force field. During the presentation of the emotional pictures, various physiological parameters were recorded from participants. The results revealed a significant difference in the self-reported arousal associated with the pictures but no significant difference in the physiological measures. The behavioural findings are interpreted in terms of an effect of the haptic feedback on participants' perceived/interpreted emotional arousal. These results suggest that haptic feedback could, in the future, be used to modify participants' interpretation of their physiological states.