Yi Fei Cheng’s research while affiliated with Carnegie Mellon University and other places

... Spatial audio is central to audio augmentation, such that virtual sounds are perceived as emanating from specific locations in 3D space [56]. Humans rely on multiple sensory modalities when they engage with their environment [35,46,67], and the auditory sense remains highly significant for localization even when visual cues is limited, sometimes replacing visual information altogether (e.g., "watching" television from another room) [14,29]. Research indicates that the use of spatial audio encourages users to adopt a more active role in spatial navigation, leading to even more accurate cognitive maps [13], while simultaneously reinforcing the sense of presence in XR environments [36]. ...

Reference:

AudioMiXR: Spatial Audio Object Manipulation with 6DoF for Sound Design in Augmented Reality

... For instance, users can become giants [10], have novel body parts [125] or even adopt non-humanoid avatars [11,72]. There is also a growing attention to how audio in VR, such as changes to the avatar's voice [19,23,25,61] or the presence of footstep sounds [65] impact immersive experience. Similarly, audio is gaining attention in remote collaboration contexts [116,129]. ...

Reference:

Audio Personas: Augmenting Social Perception via Body-Anchored Audio Cues

... Since file management mostly consists of sub-tasks such as dragging files to other folders or applying specific operations (i.e., deleting, duplicating, copying, compressing) on multiple files, a multi-selection function would be more useful. Furthermore, providing an efficient method for managing multiple objects in XR becomes more crucial, as recent XR is capable of presenting multiple windows at once [9,35]. ...

Reference:

PinchCatcher: Enabling Multi-selection for Gaze+Pinch

... To realize such adaptive MR UI behaviors, recent research formulates the problem as multi-objective optimization [9,10,16,17,22,25,26,32]. User's goals are formulated as a set of objective functions and placements are selected that maximize/minimize these objectives. ...

Reference:

Adaptive 3D UI Placement in Mixed Reality Using Deep Reinforcement Learning

... People are able to choose a variety of input methods in most Augmented Reality (AR) and Virtual Reality (VR) interfaces, and so understanding user input preferences is important. Input methods such as hand gestures and handheld controllers are widely used in AR and VR [19,33,39,46,47,50]. However, user preferences may change over time. ...

Reference:

The Brain Knows What You Prefer: Using EEG to Decode AR Input Preferences

... Aliakbarian et al. [19] also used a generative model to learn the conditional distribution of full-body poses based on the knowledge from the head and hands, named FLAG, and proposed an optimized pose prior and a new approach based on conditional normalized flow to generate high-quality poses. In order to reduce the influence of the visual range of the HMD on hand interaction, Streli et al. [20] proposed HOOV, which supplements the headset information with continuous signals from a wristband to estimate the current hand position outside the visible field of view. Some other studies suggest adding additional signal sources to track pelvic motion. ...

Reference:

Progressive Inertial Poser: Progressive Real-Time Kinematic Chain Estimation for 3D Full-Body Pose from Three IMU Sensors

... Recent advances in natural input techniques, like hand gestures [14,29,60] and speech recognition [4,35,55], have enabled more intuitive and controller-free interactions. However, these methods are often limited by a lack of haptic feedback [40,47], reduced precision [11,46], and user fatigue [6,28] over extended periods. To overcome these challenges, the focus is now on developing robust input sensing techniques that can transform everyday surfaces into interactive, touch-sensitive interfaces. ...

Reference:

HaloTouch: Using IR Multi-path Interference to Support Touch Interactions With General Surfaces

... For example, options to adjust how much an agent can intervene would preserve a teams' autonomy and allow GenAI agents to assist without overstepping or overwhelming them. Another approach that leverages the flexible collaboration spaces MR allows [44] would be to allow teams to seamlessly transition between individual, subgroup, and collective interaction modes. This would enable groups to balance collective evaluation and individual agency within their problem-solving processes as they deem appropriate. ...

Reference:

Exploring Collaborative GenAI Agents in Synchronous Group Settings: Eliciting Team Perceptions and Design Considerations for the Future of Work

... By combining machine learning techniques for spatial understanding as well as object segmentation and classification (e.g., Augmented Object Intelligence [11]), our approach for transforming virtual objects relative to their 3D centroid can be extended to physical objects as well. Such a system could create a virtual replica of relevant physical objects [31] and apply diminished reality techniques [8] to remove the physical objects from view. This would allow the virtual replica objects to be transformed for each user in the same way as in the Decoupled Hands approach. ...

Reference:

Decoupled Hands: An Approach for Aligning Perspectives in Collaborative Mixed Reality