One of the most fundamental components of our daily lives is social interaction, ranging from simple activities, such as purchasing a donut in a bakery on the way to work, to complex ones, such as instructing a remote colleague how to repair a broken automobile. While we interact with others, various challenges may arise, such as miscommunication or physical interference. In a bakery, a clerk may misunderstand the donut at which a customer was pointing due to the uncertainty of their finger direction. In a repair task, a technician may remove the wrong bolt and accidentally hit another user while replacing broken parts due to unclear instructions and lack of attention while communicating with a remote advisor.
This dissertation explores techniques for supporting multi-user 3D interaction in augmented reality in a way that addresses these challenges. Augmented Reality (AR) refers to interactively overlaying geometrically registered virtual media on the real world. In particular, we address how an AR system can use overlaid graphics to assist users in referencing local objects accurately and remote objects efficiently, and prevent co-located users from physically interfering with each other. My thesis is that our techniques can provide more accurate referencing for co-located and efficient referencing for remote users and lessen interference among users.
First, we present and evaluate an AR referencing technique for shared environments that is designed to improve the accuracy with which one user (the indicator) can point out a real physical object to another user (the recipient). Our technique is intended for use in otherwise unmodeled environments in which objects in the environment, and the hand of the indicator, are interactively observed by a depth camera, and both users wear tracked see-through displays. This technique allows the indicator to bring a copy of a portion of the physical environment closer and indicate a selection in the copy. At the same time, the recipient gets to see the indicator's live interaction represented virtually in another copy that is brought closer to the recipient, and is also shown the mapping between their copy and the actual portion of the physical environment. A formal user study confirms that our technique performs significantly more accurately than comparison techniques in situations in which the participating users have sufficiently different views of the scene.
Second, we extend the idea of using a copy (virtual replica) of physical object to help a remote expert assist a local user in performing a task in the local user's environment. We develop an approach that uses Virtual Reality (VR) or AR for the remote expert, and AR for the local user. It allows the expert to create and manipulate virtual replicas of physical objects in the local environment to refer to parts of those physical objects and to indicate actions on them. The expert demonstrates actions in 3D by manipulating virtual replicas, supported by constraints and annotations. We performed a user study of a 6DOF alignment task, a key operation in many physical task domains. We compared our approach with another 3D approach that also uses virtual replicas, in which the remote expert identifies corresponding pairs of points to align on a pair of objects, and a 2D approach in which the expert uses a 2D tablet-based drawing system similar to sketching systems developed for prior work by others on remote assistance. The study shows the 3D demonstration approach to be faster than the others.
Third, we present an interference avoidance technique (Redirected Motion) intended to lessen the chance of physical interference among users with tracked hand-held displays, while minimizing their awareness that the technique is being applied. This interaction technique warps virtual space by shifting the virtual location of a user's hand-held display. We conducted a formal user study to evaluate Redirected Motion against other approaches that either modify what a user sees or hears, or restrict the interaction capabilities users have. Our study was performed using a game we developed, in which two players moved their hand-held displays rapidly in the space around a shared gameboard. Our analysis showed that Redirected Motion effectively and imperceptibly kept players further apart physically than the other techniques.
These interaction techniques were implemented using an extensible programming framework we developed for supporting a broad range of multi-user immersive AR applications. This framework, Goblin XNA, integrates a 3D scene graph with support for 6DOF tracking, rigid body physics simulation, networking, shaders, particle systems, and 2D user interface primitives.
In summary, we showed that our referencing approaches can enhance multi-user AR by improving accuracy for co-located users and increasing efficiency for remote users. In addition, we demonstrated that our interference-avoidance approach can lessen the chance of unwanted physical interference between co-located users, without their being aware of its use.
| Identifer | oai:union.ndltd.org:columbia.edu/oai:academiccommons.columbia.edu:10.7916/D8BC3ZCK |
| Date | January 2016 |
| Creators | Oda, Ohan |
| Source Sets | Columbia University |
| Language | English |
| Detected Language | English |
| Type | Theses |
Page generated in 0.0019 seconds