arXiv subjects

Misha Sra

Publications and source records attributed to Misha Sra.

At least 19 recordsLinked to original sources

arXiv2026

The Missing Link of XR: Empathy-Driven Reality for XR and Beyond

Extended reality (XR) for socialising is becoming increasingly popular. However, unlike conventional social platforms, XR prioritises embodiment and immersion, factors that strongly impact one's physical and mental states. We envision a future for XR where all users, regardless of abilities and backgrounds, can understand one another, participate, and find safe socialisation spaces. An Empathy-Driven Reality (EDR) is a space where understanding each other's emotional, physical, and cognitive states takes centre stage. It has the potential to enhance empathy beyond how we normally perceive it. To explore this concept, we conducted a hybrid-style workshop over two months with 27 industry and academic researchers in XR, emotion, physiology, assistive technology, and social science. This paper reports on the findings and aims to establish a structure and reference for the 1) design guidelines, 2) research challenges, and 3) potential applications for the future of XR as an EDR. We frame EDR as a conceptual design framework rather than a validated system: a structured design space that links requirements to design mechanisms, together with a research agenda for empathy-centred XR.

cs.HC↗

arXiv2026

AutoLab: Can Frontier Models Solve Long-Horizon Auto Research and Engineering Tasks?

Scientific and engineering progress is fundamentally a long-horizon iterative process: proposing changes, running experiments, measuring outcomes, and continuously refining artifacts. Yet existing benchmarks for frontier models primarily evaluate either single-turn responses or short-horizon agent trajectories, failing to capture the challenges of sustained iterative improvement over extended time horizons. To address this gap, we introduce AutoLab, a new benchmark for ultra long-horizon closed-loop optimization. AutoLab consists of 36 realistic, expert-curated tasks spanning four diverse domains: system optimization, puzzle & challenge, model development, and CUDA kernel optimization. Each task begins with a correct but deliberately suboptimal baseline and challenges agents to improve it within a strict wall-clock budget. Evaluating 17 state-of-the-art models reveals the dominant predictor of success is not the quality of an agent's initial attempt, but its persistence in repeatedly benchmarking, editing, and incorporating empirical feedback. While claude-opus-4.6 exhibits strong long-horizon optimization capabilities, most frontier models, including several proprietary ones, either terminate prematurely or exhaust their budgets with minimal progress. These results underscore the importance of time awareness and persistent iteration in autonomous agents. We open-source the full benchmark, evaluation harness, and task artifacts, to accelerate research toward truly capable long-horizon agents.

cs.AI↗

arXiv2026

Visual Aesthetic Benchmark: Can Frontier Models Judge Beauty?

Multimodal large language models (MLLMs) are now routinely deployed for visual understanding, generation, and curation. A substantial fraction of these applications require an explicit aesthetic judgment. Most existing solutions reduce this judgment to predicting a scalar score for a single image. We first ask whether such scores faithfully capture comparative preference: in a controlled study with eight expert annotators, score-derived rankings align poorly with the same annotators' direct comparisons, while direct ranking yields substantially higher inter-annotator agreement on best- and worst-image labels. Motivated by this finding, we introduce the Visual Aesthetic Benchmark (VAB), which casts aesthetic evaluation as comparative selection over candidate sets with matched subject matter. VAB contains 400 tasks and 1,195 images across fine art, photography, and illustration, with labels derived from the consensus of 10 independent expert judges per task. Evaluating 20 frontier MLLMs and six dedicated visual-quality reward models, we find that the strongest system identifies both the best and the worst image correctly across three random permutations of the candidate order in only 26.5% of tasks, far below the 68.9% achieved by human experts. Fine-tuning a 35B-parameter model on 2,000 expert examples brings its accuracy close to that of a 397B-parameter open-weight model, suggesting that the comparative signal in VAB is transferable. Together, these results expose a clear and measurable gap between current multimodal models and expert aesthetic judgment, and VAB provides the first set-based, expert-grounded testbed on which that gap can be tracked and closed.

cs.CV↗

arXiv2026

MiXR: Harvesting and Recomposing Geometry from Real-World Objects for In-Situ 3D Design

Recent developments in 3D generative AI enable users to create bespoke 3D models from text or image prompts. However, these approaches provide limited control over spatial structure, making them ill suited for tasks requiring precise geometric composition. We present MiXR, an XR system for in-situ compositional modeling that enables users to create new 3D models by harvesting geometry from their environment. Users extract segments from captured objects and assemble new artifacts through direct 3D manipulation, while generative AI synthesizes a coherent model from the user-defined composition. This hybrid workflow allows users to define spatial structure explicitly while delegating geometric refinement to generative models, enabling them to specify spatial intent that is difficult to express through verbal prompts alone. In a controlled user study ($N=12$), participants using MiXR rated their designs as significantly closer to the target, felt more in control, and experienced lower cognitive workload compared to a generative composition baseline.

cs.HC↗

arXiv2026

Exploration of Radar-based Obstacle Visualizations to Support Safety and Presence in Camera-Free Outdoor VR

Outdoor virtual reality (VR) places users in dynamic physical environments where they must remain aware of real-world obstacles, including static structures and moving bystanders, while immersed in a virtual scene. This dual demand introduces challenges for both user safety and presence. Millimeter-wave (mmWave) radar offers a privacy-preserving alternative to camera-based sensing by detecting obstacles without capturing identifiable visual imagery, yet effective methods for communicating its sparse spatial information to users remain underexplored. In this work, we developed and validated WaveWalkerClone, a reproduction of the WaveWalker system, to establish reliable radar- and GPS-IMU-based sensing under varied outdoor lighting conditions. Building on this feasibility validation, we conducted a user study (n=18) comparing three visualization techniques for radar-detected obstacles : (1) diegetic alien avatars that visually embed obstacles within the virtual narrative, (2) non-diegetic human avatars represented obstacles as humans inconsistent with the virtual narrative, and (3) abstract point clouds centered around the obstacles conveying spatial data without anthropomorphic or narrative associations. Our results show that all three approaches supported user safety and situational awareness, but yielded distinct trade-offs in perceived effort, frustration, and user preference. Qualitative feedback further revealed divergent user responses across conditions, highlighting the limitations of a one-size-fits-all approach. We conclude with design considerations for obstacle visualization in outdoor VR systems that seek to balance immersion, safety, and bystander privacy.

cs.HC↗

arXiv2026

How Users Perceive Mixed-Initiative AI: Attitudes Toward Assistance in Problem Solving

In mixed-initiative systems, the mode of AI assistance delivery can be as consequential as the assistance itself. We investigated two assistance delivery modes: on-demand help (users request via Button) and pre-scheduled help (assistance delivered at user-selected intervals, with user actions resetting the Timer). To evaluate these modes, we selected Rush Hour puzzles as the human-AI collaborative task because they capture elements of real-world problem solving such as analysis, resource management, and decision-making under constraints. To enhance ecological validity, we imposed monetary costs for both time and AI assistance, simulating scenarios where people must balance implicit or explicit trade-offs such as time pressure, financial limitations, or opportunity costs. Although task performance was comparable across modes, participants who used the pre-scheduled (Timer) mode reported more positive perceptions of the AI, even when their ending budget was low. This suggests that assistance delivery mode can shape user experience independent of task outcomes, indicating that human-AI systems may need to consider how AI assistance is delivered alongside improving task performance.

cs.HC↗

arXiv2026

Embedded vs. Situated: An Evaluation of AR Facial Training Feedback

While augmented reality (AR) research demonstrates benefits of embedded visualizations for gross motor training, its applicability to facial exercises remains under-explored. Providing effective real-time feedback for facial muscle training presents unique design challenges, given the complexity of facial musculature. We developed three AR feedback approaches varying in spatial relationship to the user: situated (screen-fixed), proxy-embedded (on a mannequin), and fully embedded (overlaid on the user's face). In a within-subjects study (N=24), we measured exercise accuracy, cognitive load, and user preference during facial training tasks. The embedded feedback reduced cognitive load and received higher preference ratings, while the situated feedback enabled more precise corrections and higher accuracy. Qualitative analysis revealed a key design tension: embedded feedback improved experience but created self-consciousness and interpretive difficulty. We distill these insights into design considerations addressing the trade-offs for facial training systems, with implications for rehabilitation, performance training, and motor skill acquisition.

cs.HC↗

arXiv2026

Grand Challenges around Designing Computers' Control Over Our Bodies

Advances in emerging technologies, such as on-body mechanical actuators and electrical muscle stimulation, have allowed computers to take control over our bodies. This presents opportunities as well as challenges, raising fundamental questions about agency and the role of our bodies when interacting with technology. To advance this research field as a whole, we brought together expert perspectives in a week-long seminar to articulate the grand challenges that should be tackled when it comes to the design of computers' control over our bodies. These grand challenges span technical, design, user, and ethical aspects. By articulating these grand challenges, we aim to begin initiating a research agenda that positions bodily control not only as a technical feature but as a central, experiential, and ethical concern for future human-computer interaction endeavors.

cs.HC↗

arXiv2026

AlignUI: A Method for Designing LLM-Generated UIs Aligned with User Preferences

Designing user interfaces that align with user preferences is a time-consuming process, which requires iterative cycles of prototyping, user testing, and refinement. Recent advancements in LLM-based UI generation have enabled efficient UI generation to assist the UI design process. We introduce AlignUI, a method that aligns LLM-generated UIs with user tasks and preferences by using a user preference dataset to guide the LLM's reasoning process. The dataset was crowdsourced from 50 general users (the target users of generated UIs) and contained 720 UI control preferences on eight image-editing tasks. We evaluated AlignUI by generating UIs for six unseen tasks and conducting a user study with 72 additional general users. The results showed that the generated UIs closely align with multiple dimensions of user preferences. We conclude by discussing the applicability of our method to support user-aligned UI design for multiple task domains and user groups, as well as personalized user needs.

cs.HC↗

arXiv2025

TaskLens: Generating Task-Conditioned Scaffolded Interfaces for Learning Professional Creative Software

Professional creative software has steep learning curves for novices due to complex interfaces, limited guidance, and unfamiliar terminology. To support educators and tool creators in addressing learner challenges, we introduce TaskLens, an LLM-based method that automatically generates task-conditioned scaffolded UIs from natural language task descriptions. Our method uses LLMs to identify workflow stages and domain concepts, select task-relevant tools, generate implementation code, and execute the code to produce scaffolded interfaces. The interfaces surface relevant tools, organize them by workflow stage, link them to domain concepts, and progressively disclose advanced features. We evaluate TaskLens by deploying two LLM-generated scaffolded interfaces in Blender, a professional 3D modeling software. A user study with beginners (n=32) showed that our scaffolded interfaces significantly reduced perceived task load, improved task performance through embedded workflow guidance, and increased domain concept learning in Blender during task execution. A second study with experts (n=8) showed improved task efficiency and potential to create personalized UIs for productivity and creativity.

cs.HC↗

arXiv2025

Audience Amplified: Virtual Audiences in Asynchronously Performed AR Theater

Audience reactions can considerably enhance live experiences; conversely, in anytime-anywhere augmented reality (AR) experiences, large crowds of people might not always be available to congregate. To get closer to simulating live events with large audiences, we created a mobile AR experience where users can wander around naturally and engage in AR theater with virtual audiences trained from real audiences using imitation learning. This allows us to carefully capture the essence of human imperfections and behavior in artificial intelligence (AI) audiences. The result is a novel mobile AR experience in which solitary AR users experience an augmented performance in a physical space with a virtual audience. Virtual dancers emerge from the surroundings, accompanied by a digitally simulated audience, to provide a community experience akin to immersive theater. In a pilot study, simulated human avatars were vastly preferred over just audience audio commentary. We subsequently engaged 20 participants as attendees of an AR dance performance, comparing a no-audience condition with a simulated audience of six onlookers. Through questionnaires and experience reports, we investigated user reactions and behavior. Our results demonstrate that the presence of virtual audience members caused attendees to perceive the performance as a social experience with increased interest and involvement in the event. On the other hand, for some attendees, the dance performances without the virtual audience evoked a stronger positive sentiment.

cs.HC↗

arXiv2025

ReUseIt: Synthesizing Reusable AI Agent Workflows for Web Automation

AI-powered web agents have the potential to automate repetitive tasks, such as form filling, information retrieval, and scheduling, but they struggle to reliably execute these tasks without human intervention, requiring users to provide detailed guidance during every run. We address this limitation by automatically synthesizing reusable workflows from an agent's successful and failed attempts. These workflows incorporate execution guards that help agents detect and fix errors while keeping users informed of progress and issues. Our approach enables agents to successfully complete repetitive tasks of the same type with minimal user intervention, increasing the success rates from 24.2% to 70.1% across fifteen tasks. To evaluate this approach, we invited nine users and found that our agent helped them complete web tasks with a higher success rate and less guidance compared to two baseline methods, as well as allowed users to easily monitor agent behavior and understand its failures.

cs.HC↗

arXiv2025

CoDA: Agentic Systems for Collaborative Data Visualization

Deep research has revolutionized data analysis, yet data scientists still devote substantial time to manually crafting visualizations, highlighting the need for robust automation from natural language queries. However, current systems struggle with complex datasets containing multiple files and iterative refinement. Existing approaches, including simple single- or multi-agent systems, often oversimplify the task, focusing on initial query parsing while failing to robustly manage data complexity, code errors, or final visualization quality. In this paper, we reframe this challenge as a collaborative multi-agent problem. We introduce CoDA, a multi-agent system that employs specialized LLM agents for metadata analysis, task planning, code generation, and self-reflection. We formalize this pipeline, demonstrating how metadata-focused analysis bypasses token limits and quality-driven refinement ensures robustness. Extensive evaluations show CoDA achieves substantial gains in the overall score, outperforming competitive baselines by up to 41.5%. This work demonstrates that the future of visualization automation lies not in isolated code generation but in integrated, collaborative agentic workflows.

cs.AI↗

arXiv2025

Understanding Mode Switching in Human-AI Collaboration: Behavioral Insights and Predictive Modeling

Human-AI collaboration is typically offered in one of two of user control levels: guidance, where the AI provides suggestions and the human makes the final decision, and delegation, where the AI acts autonomously within user-defined constraints. Systems that integrate both modes, common in robotic surgery or driving assistance, often overlook shifts in user preferences within a task in response to factors like evolving trust, decision complexity, and perceived control. In this work, we investigate how users dynamically switch between higher and lower levels of control during a sequential decision-making task. Using a hand-and-brain chess setup, participants either selected a piece and the AI decided how it moved (brain mode), or the AI selected a piece and the participant decided how it moved (hand mode). We collected over 400 mode-switching decisions from eight participants, along with gaze, emotional state, and subtask difficulty data. Statistical analysis revealed significant differences in gaze patterns and subtask complexity prior to a switch and in the quality of the subsequent move. Based on these results, we engineered behavioral and task-specific features to train a lightweight model that predicted control level switches ($F1 = 0.65$). The model performance suggests that real-time behavioral signals can serve as a complementary input alongside system-driven mode-switching mechanisms currently used. We complement our quantitative results with qualitative factors that influence switching including perceived AI ability, decision complexity, and level of control, identified from post-game interview analysis. The combined behavioral and modeling insights can help inform the design of shared autonomy systems that need dynamic, subtask-level control switches aligned with user intent and evolving task demands.

cs.HC↗

arXiv2025

XARP Tools: An Extended Reality Platform for Humans and AI Agents

Building XR-AI research prototypes requires navigating two largely separate ecosystems. Mainstream XR development relies on C#/C++ and game engines, while AI development is centered on Python. This toolchain fragmentation slows down contributions to human-AI spatial interaction research. To broaden access to XR development in the Python ecosystem, we present XARP (XR Agent-ready Remote Procedures), a toolkit for rapid XR-AI prototyping in Python. XARP application logic runs on a Python server and controls a Unity client through WebSocket messages. This architecture enables compatibility with multiple client platforms and live reloading of application code without client redeployment. XARP is available to humans as a library and to AI agents as callable tools and through Model Context Protocol. We designed XARP through formative case studies and refined it through an early acceptance evaluation with 24 XR and AI developers and a six-week longitudinal study with two developers building an independent research project. Potential users expected the toolkit to improve their performance and facilitate development. Sustained use confirmed faster iteration and easier setup compared to conventional XR workflows, with asset-intensive and performance-critical projects emerging as the clearest limitations. Technical benchmarks show that hand and head tracking data streaming was close to the device refresh rate of 72 FPS, and that AI agents using XARP consumed 19% fewer tokens than those writing equivalent C# Unity code. Beyond broadening access to XR development, XARP reduces engineering friction in spatial computing research and opens new pathways for AI agents to participate in XR application development. XARP is open source and available at https://github.com/hal-ucsb/xarp.

cs.HC↗

arXiv2025

Designing Scaffolded Interfaces for Enhanced Learning and Performance in Professional Software

Professional software offers immense power but also presents significant learning challenges. Its complex interfaces, as well as insufficient built-in structured guidance and unfamiliar terminology, often make newcomers struggle with task completion. To address these challenges, we introduce ScaffoldUI, a method for scaffolded interface design to reduce interface complexity, provide structured guidance, and enhance software learnability. The scaffolded interface presents task-relevant tools, progressively discloses tool complexity, and organizes tools based on domain concepts, aiming to assist task performance and software learning. To evaluate the feasibility of our interface design method, we present a technical pipeline for scaffolded interface implementation in professional 3D software, i.e., Blender, and conduct user studies with beginners (N=32) and experts (N=8). Study results demonstrate that our scaffolded interfaces significantly reduce perceived task load caused by interface complexity, support task performance through structured guidance, and augment learning by clearly connecting concepts and tools within the taskflow context. Based on a discussion of the user study findings, we offer insights for future research on designing scaffolded interfaces to support instruction, productivity, creativity, and cross-software workflows.

cs.HC↗

arXiv2025

Instruct-CLIP: Improving Instruction-Guided Image Editing with Automated Data Refinement Using Contrastive Learning

Although natural language instructions offer an intuitive way to guide automated image editing, deep-learning models often struggle to achieve high-quality results, largely due to the difficulty of creating large, high-quality training datasets. To do this, previous approaches have typically relied on text-to-image (T2I) generative models to produce pairs of original and edited images that simulate the input/output of an instruction-guided image-editing model. However, these image pairs often fail to align with the specified edit instructions due to the limitations of T2I models, which negatively impacts models trained on such datasets. To address this, we present Instruct-CLIP (I-CLIP), a selfsupervised method that learns the semantic changes between original and edited images to refine and better align the instructions in existing datasets. Furthermore, we adapt Instruct-CLIP to handle noisy latent images and diffusion timesteps so that it can be used to train latent diffusion models (LDMs) and efficiently enforce alignment between the edit instruction and the image changes in latent space at any step of the diffusion pipeline. We use Instruct-CLIP to correct the InstructPix2Pix dataset and get over 120K refined samples we then use to fine-tune their model, guided by our novel I-CLIP-based loss function. The resulting model can produce edits that are more aligned with the given instructions. Our code and dataset are available at https://github.com/SherryXTChen/Instruct-CLIP.git.

cs.CV↗

arXiv2025

Standard Benchmarks Fail -- Auditing LLM Agents in Finance Must Prioritize Risk

Standard benchmarks fixate on how well large language model (LLM) agents perform in finance, yet say little about whether they are safe to deploy. We argue that accuracy metrics and return-based scores provide an illusion of reliability, overlooking vulnerabilities such as hallucinated facts, stale data, and adversarial prompt manipulation. We take a firm position: financial LLM agents should be evaluated first and foremost on their risk profile, not on their point-estimate performance. Drawing on risk-engineering principles, we outline a three-level agenda: model, workflow, and system, for stress-testing LLM agents under realistic failure modes. To illustrate why this shift is urgent, we audit six API-based and open-weights LLM agents on three high-impact tasks and uncover hidden weaknesses that conventional benchmarks miss. We conclude with actionable recommendations for researchers, practitioners, and regulators: audit risk-aware metrics in future studies, publish stress scenarios alongside datasets, and treat ``safety budget'' as a primary success criterion. Only by redefining what ``good'' looks like can the community responsibly advance AI-driven finance.

q-fin.GN↗