If you have any problems related to the accessibility of any content (or if you want to request that a specific publication be accessible), please contact (email@example.com). We will work to respond to each request in as timely a manner as possible.
Intent Recognition Using an Activation Spreading Architecture
AuthorSaffar, Mohammad Taghi
Computer Science and Engineering
StatisticsView Usage Statistics
Intent understanding is the problem of recognizing people’s goals by passively observing them perform some activities and predicting their future actions. In this thesis, we designed and implemented an efficient intent recognition system with real-time capabilities for two different applications. One of the applications is to recognize intentions in highly populated multi-agent environments with real-time constraints. Low-level intention between all pairs of agents are detected through a novel formulation of Hidden Markov Models with perspective taking capabilities. This layer of our framework only captures interactions between pairs of agents. To move to a higher level and also detect intentions involving multiple agents we use a distributed representation of connecting intentions based on the idea of Activation Spreading Networks (ASN). We utilized an open source naval ship simulator and showed that our system is able to detect intentions reliably and early. Another application is to recognize intentions of humans by observing their activities with an RGB-D camera. Activities and goals are modeled as a distributed network of inter-connected nodes in an ASN. Inspired by a formalism in hierarchical task networks, the structure of the network captures the hierarchical relationship between high-level goals and low-level activities that realize these goals. Our approach can detect intentions before they are realized and it can work in real-time. We also extend the formalism of ASNs to incorporate contextual information into intent recognition. We further augment the ASN formalism with special nodes and synaptic connections to model ordering constraints between actions, in order to represent and handle partial-order plans in our ASN.