1Human-Computer Interaction Collaborative Discourse and Plan RecognitionThomas CoffeeShannon DongShen Qu4/13/2005Through this lecture, we aim to convey the usefulness of intelligent systems capable of understanding and collaborating with humans towards the accomplishment of a common goal. Specifically, we will focus on the theory of collaborative discourse and plan recognition and their role in current collaborative assistants.22Hello, Computer?“Hello, computer?” These were the words uttered by Professor Scotts (a.k.a. Scotty) to the mouse of a 20thcentury computer when the crew of the Enterprise traveled back in time to year 1986. This scene is probably a familiar sight to all Star Trek fans, but one may wonder “what does it have to do with cognitive robotics?”Well, anyone who’s seen any form of Star Trek probably encountered scenario(s) where a crewmember works with the onboard computer system to solve various types of complex problems. And although not immediately obvious, this problem solving process not only involves command inputs from the human party but also demands a computer software system capable of keeping track of the problem’s state, any progress made, and what still needs to be done while providing inputs and quires as needed. When Scotty said “hello computer,” he was not simply seeking a “how do you do” reply but was trying to start a collaborative session in which the computer would assist him in the accomplishment of a goal (in this case, constructing the molecular structure of transparent aluminum).Through this lecture, we will demonstrate the importance/usefulness of such computer/software agents, present a couple of such agents currently under development, and dive into the theory behind a couple of key building blocks of these agents.33TRIPS DemoBefore going into technical details, we would first like to show a demonstration of a collaborative agent TRIPS applied to a military scenario. The demo is self explanatory in terms of the problem statement and final goals. We are showing this demo early on to give people a sense of what is available today.Note that in this scenario the computer received a high level goal from the user, provided relevant information without being specifically prompted to do so, made recommendations towards the course of action, kept track of parallel solution approaches to the goal, retraced steps when an approach did not work out, asked for clarification when information provided is unclear or incomplete, and provided both visual and vocal feedback to the user. These are all important properties for this type of collaborative agents, and towards the end of this lecture we will show a similar demo in which we link these key features to the theory that we will soon present.This demo can be found at the following url: http://www.cs.rochester.edu/research/cisd/projects/trips/movies/TRIPS_CPoF/44How can we create computer or robotic systems that not only follow user commands but also collaborate with humans toward the achievement of a common goal?This question outlines the main objective of this lecture. Key words to note are “collaborate” and “common goal.”55Technical Approach• Derive intentions from observed or communicated information• Decompose goals into organized structure• Refine goal structures by seeking additional information• Identify shared elements of agents’ goals• Execute communications and actions to achieve multi-agent objectivesHuman-computer/robot interaction is a wide field. Even a specific type of interactive agent such as TRIPS is very complex. This slide only outlines the technical approach to what we consider to be the most important components to this type of agents.66Outline• Travel Scheduling Example• Collaborative Discourse• Plan Recognition• TRIPS• COLLAGENFor the reminder of this lecture, we will first introduce a travel planning example in which collaborative agents can be applied. And we will continue to refer back to this example through our technical discussion on Collaborative Discourse and Plan Recognition. Finally we will end the lecture with a more detailed analysis of how the 2 leading collaborative agents TRIPS and COLLAGEN embody the theory presented along with a demo for each agent.77Air Travel Scheduling ScenarioSan FranciscoBostonDallasPhoenixLeave: Tuesday night or Wednesday morningReturn: Friday before 5pmThursday 11am-3pmPrefer to fly American Airlines because of frequent flier milesMany people are familiar with the complexity and headaches associated with travel scheduling and ticket purchasing. Above is a scenario where a Boston based traveler wants to make a trip to Dallas, Phoenix, and San Francisco. There are some hard constraints such as he must be in Phoenix on Thursday between 11am and 3pm and return to Boston by Friday 5pm. There are also lose constraints and preferences such as he would like to leave Wednesday morning but can leave Tuesday night, and he would like to fly American Airlines as much as possible (but this is not a requirement).88If he visits a website such as Yahoo! Travel to plan his trip, one can immediately see the problem. The simple “to” and “from” fields along with date constraints as no where near sophisticated enough to solve this multi-destination trip with varying levels of constraints.99Problems with Travel Planning• Order of actions may not be flexible• Difficult to recover from mistakes• Easy to get stuck or get lost• May over or under constrain• Lack of support for the user’s problem solving process as it unfolds over timeHere are some of the common problems with travel planning systems, or even planners and schedulers in general.1) Order of actions may not be flexible: The user may be required to enter the destination before the date and time information.2) Difficult to recover from mistakes: Due to the inflexibility of actions changing the destination may erase other data already entered.3) Easy to get stuck or get lost: After trying 20 or 30 combinations of departure and destination location and time, various airlines, perhaps even multiple websites, it would be very easy to loose track of the combinations that have been tried and the possible itineraries listed.4) May over or under constrain: One can easily over or under constrain the problem and receive 100 possible itineraries or no itinerary at all.5) And the main reason for all this confusion is that
View Full Document
Unlocking...