Keru Wang

We designed a proactive AI agent that monitors group brainstorming and intervenes at opportune moments to counter idea stagnation. Through two user studies with 18 participants, we found that the proactive agent boosted idea generation, depth, and novelty compared to a reactive baseline, though users still preferred reactive support for its greater sense of control.

We explore how proactive AI agents can step in at the right moments to keep group brainstorming productive and natural. Using a custom LLM system that monitors discussions and explains its reasoning, we compared its timing judgement to expert evaluations. The study uncovers what makes AI interventions feel helpful versus disruptive, guiding the design of more intuitive and collaborative AI partners.

We developed a VR-based shared control framework that allows operators to guide teams of drones in complex, unknown environments. A motion-primitive planner integrates real-time user input to generate collision-free trajectories, while an admittance controller enables flexible human influence. Experiments show that this immersive, human-in-the-loop approach enhances obstacle avoidance, coordination, and operator efficiency.

This survey explores how auditory cues can evoke pseudo-haptic sensations, offering a cost-effective alternative to traditional haptic devices. We review existing research on audio-influenced pseudo-haptics, summarize mappings between sound and touch perception, and outline future opportunities for creating more immersive and accessible interactive systems.

Robotecture is a scalable, cost-efficient shape-changing system using modular actuated beams to create dynamic surfaces and enclosures. Its flexible, space-efficient design enables applications ranging from tabletop displays and smart storage systems to smart architecture.

We present a VR-based terrain generation system that uses hand gestures and a generative model for fast landscape prototyping. Users draw and modify mid-air strokes to outline shapes, and a Conditional GAN, trained on real terrains' height maps, generates realistic landscapes blending user input with training features.

Our collaborative XR system integrates physical and virtual design spaces, superimposing visual modifications on physical objects via video passthrough. With multimodal inputs, real-time object tracking, and 3D annotations, it accelerates design iterations for digital prototyping.

The system lets users intuitively use 'object-level' instructions to control a multi-robot setup to manipulate tabletop objects via gestures, GUI, tangible inputs, and speech. Robots collectively execute commands through a generalizable algorithm, enabling high-level, object-focused multimodal interactions.

This paper presents the ongoing development of a spatial audio system for co-located, multi-participant, extended reality (CMXR) experiences. By integrating spatial audio and informative auditory displays, the system can enhance the sense of immersion and presence among participants and facilitate collaboration.

We developed a system that allows non-VR instructors to use hand-held devices, like smartphones and tablets, to explore VR content and interact with fully immersed students. Instructors can observe the VR environment, switch between students’ views, and interact within the VR world, while students view real-time video streams of both the physical space and the instructor.

Our project combines immersive VR, multitouch AR, real-time volumetric capture, motion capture, robotically-actuated tangible interfaces at multiple scales, and live coding, in service of a human-centric way of collaborating.

GazeChat is a remote conferencing system that uses gaze-aware 3D profile photos to enrich online conversations. Through webcam-based gaze tracking and neural rendering, GazeChat redirects participants' gaze to accurately reflect who they’re looking at, enhancing communication and engagement.