Lanny Lin

From HCMI

Jump to: navigation, search
The real Lanny, not a robot.

I am a PhD student in the Computer Science Department at BYU. I received my undergraduate degrees from Southern Oregon University in 1997. After working in the industry for 9 years, I decided to return to school to pursue something that I've always wanted to do, something I could enjoy for the rest of my life --- Artificial Intelligence and Robotics --- hence, here I am.

Contents

Research Interests

I am interested in how an end user can better manage a robot's autonomous behaviors and teach the robot to perform new tasks. This is a human-centered approach but also requires machine autonomy, and of course, there are the human-robot interaction interfaces.

I believe that human can better manage autonomy by managing information provided to an intelligent system at different scales and resolutions.

  • At the highest scale (lowest resolution), the intelligent system can model the world and then predict general trends. The human operator can determine what information to provide to the intelligent system and what parameters to use in the world model.
  • At the medium scale (medium resolution), the human operator can perform case-specific planning and then provide the intelligent system additional information such as areas of focus/interest.
  • At the lowest scale (highest resolution), the human operator can manage autonomous behaviors at the execution level by specifying areas of focus/interest in real-time. Additional information, such as task difficulty and desired intensity can also be specified by the human user as an indirectly means to affect the intelligent system's autonomous behaviors.

To test out the proposed approach, I am applying it to two different application domains:

  • Using Unmanned Aerial Vehicles to support Wilderness Search and Rescue.
  • Using humanoid robots to assist therapists in treating children with autism in clinical settings.


Research Groups

I am currently involved in two research groups:

  • WiSAR Group -- Using UAV to support Wilderness Search and Rescue
  • TiLAR Group -- Therapist-in-the-loop Assistive Robotics


Research Projects

Upper: Prior predictive probability distribution suggested by the Bayesian model. Lower: Posterior predictive probability distribution.
Left: A sample multi-modal probability distribution of the likely places to find the missing person with desired starting point marked. Right: Search path produced by the intelligent path planning algorithms.
A mock up UAV search path showing human planning strategically (by specifying way points) and intelligent algorithms planning tactically (maximize amount of probability collected given a probability distribution map).
  • Modeling Terrain-based Lost Person Behavior -- Using publicly available terrain data, we built a Bayesian model to predict how terrain features (such as topography, vegetation density and elevation difference) might affect a lost person's behavior. This model allows the searcher to incorporate his/her uncertainly in the prior estimates of lost person's transitional probabilities between two terrain features. It also allows the searcher to incorporate past human behavior data in the form of GPS track logs to generate posterior predictive probability distribution. We are presently working on extending this model to include intended destination and trail following factors and then use geocacher GPS track logs as observed human behavior data in our model. We are also developing user interfaces that allow the searcher to easily view and modify transitional probabilities in the form of Beta Distributions.


  • Gesture-based Probability Distribution Modification Tool -- Once the system produces a probability distribution of the likely places to find the missing person based on general trend prediction, the searcher needs to be able to modify the probability distribution based on the very specific case scenario at hand. The searcher would have better information about the profile of the missing person and other additional information such as the season of the year, time of the day, and weather conditions. The tool we are building would allow the searcher to use simple gestures (eventually using a touch-screen monitor) to modify the suggested probability distribution by raising, lowering, and erasing probability distribution hills. This tool is not only useful in the planning phase, it can also be useful in the execution phase when new evidences are found that might change the probability distribution.


  • Intelligent Path Planning Algorithms -- Give a probability distribution of the likely places of finding the missing person, what path should a UAV follow to maximize the amount of probability collected? We have developed path planning algorithms that combine various techniques (such as Local Hill Climbing, Potential Fields, Global Warming Effect, and Evolutionary Algorithms) to produce a path from specified starting position (to an optional ending position) for a fixed flight duration. We tested the algorithms on various type of probability distributions (uni-modal, bi-modal, multi-modal, uniform, and path) and were able to plan paths that exceed 95% efficiency. We are currently working on speeding up the path planning by performing coarse-to-fine search in the Global Warming dimension. We are also investigating how well our algorithms work with partial detection rate and also the additional time dimension (planning path for a moving object/changing distribution).


  • Sliding Autonomy Intensity Control -- During the execution phase when using a UAV to support Wilderness Search and Rescue, the searcher can plan the UAV search path strategically by specify way points in different region of the search area. Intelligent path planning algorithms are then used to plan flying path tactically between two consecutive way points based on the given probability distribution of the likely places to find the missing person. The searcher can using a slider to control how much time to grant the UAV between two consecutive way points. This controlling mechanism enables the searcher to control the search intensity at the local region. Real-time feedback allows the searcher to see what search paths are generated and then select the path he/she likes the most.


  • Probability of Detection Map Painting Tool -- Although the UAV can provide aerial video support, if an object is in the video frames, it doesn't mean the searcher will always spot it due to various factors such as vegetation density, weather conditions, and fatigue. This tool allows the searcher to paint a probability of detection map for the search region marking areas where probability of detection might be low. This way the searcher can provide more information to the intelligent system, and the intelligent path planning algorithms can take the probability of detection factor into account accordingly when planning flight paths for the UAV.


  • Autonomy Management Tool for Therapists -- When treating children with autism, the therapists might like to focus on different areas of treatment for different children or at different sessions for the same child. Examples include verbal communication, turn taking, joint attention, and eye contacts. This proposed tool allows a therapist to specify areas of interest for a specific kid in a specific session and then use slider controls to control the (semi)autonomous interactions among the therapist, the child, and the robot both in the planning phase and the execution phase (during the session) by controlling: 1) the exaggeration effect (facial expression and body movement/gesture), 2) the verbose level, and 3) the robot's reliance on the therapist.


Other Projects Interested

  • Common object identification with web-based knowledge extraction -- If I ask a robot to get me a soccer ball, the robot should be able to quickly learn from the Internet what a soccer ball looks like and what are some of the physical attributes of a soccer ball (e.g., it is a sphere-shaped object) and where are likely places to find it (e.g., in Wal-Mart, the Gym, the closet, etc.). Human can then help the robot learn by providing praises or corrections.
  • How to make a robot rap and dance -- Software today can already automatically analyze the rhythms of songs and speak in synthetic voice using text-to-speech technologies. Combining that with search engine technology, we should be able to automatically generate rap lyrics given a topic and a piece of music. Some of the challenges include intonation control and speed control with text-to-speech, so the singing matches well with the rhythms and sentences are complete in the lyrics. Other challenges include applying patterns to the lyrics so it sounds more like a song instead of endless nabbing. A robot that's capable of detecting its own body parts (external moving parts, that is) can then dance to the music and become a robot rapper.
  • Assistive Robotics for Eldercare -- A home assistant robot can help monitor elders in independent living autonomously by checking on the elders periodically or on demand (from remote relatives). The robot should be able to take vital signs from the elder (with the elder's consent of course) and then automatically send the collected data to registered authorities (e.g., a hospital). It should be able to alarm relatives or the authority when it observes abnormal conditions (elder falling or lying unconsciously on the floor). It should also be able to periodically check around to make sure certain appliances are turned off (e.g., stove, TV, etc.) at night.
  • Robot Companionship for Elders -- A companion robot should be able to support tele-presence from relatives so the elder can talk to it like talking to the real person (the monitor will show the caller's face like in video conferencing, the robot speakers will become phone speakers, and the robot/phone will go to the elder instead of the elder going to the robot/phone, and the robot can follow the elder around). As an autonomous companion, the robot should be able to read news or jokes to the elder (from the Internet) and entertain the elder (by conversing with the elder, or sing/dance, or whatever else the robot is capable of). It should also be able to provide friendly reminders such as time for medication or birthday's of grandchildren, or doctor's appointment, etc.


Personal Interests

I have too many hobbies. Some of the leading ones are soccer, billiards, music improvisation, and martial arts. In my spare time (if I can squeeze out any), I translate Chinese martial arts novels into English and also maintain a personal blog because I believe that good things should be shared. Welcome to check them out:


Publications

L. Lin and M. A. Goodrich. A Bayesian Approach to Modeling Lost Person Behaviors Based on Terrain Features in Wilderness Search and Rescue. To appear in Computational and Mathematical Organization Theory 2010 Special Issue.

L. Lin, M. Roscheck, M. A. Goodrich, and B. S. Morse. Supporting Wilderness Search and Rescue with Integrated Intelligence: Autonomy and Information at the Right Time and the Right Place . To appear in Twenty-Fourth AAAI Conference on Artificial Intelligence, Special Track on Integrated Intelligence. July, 2010, Atlanta, Georgia, USA.

L. Lin and M. A. Goodrich. A Behavior-based Interactive Learning Approach for HRI: Teaching a Robot New Tricks. In Technical Report of the HRI 2010 Young Pioneers Workshop. Osaka, Japan. March, 2010.

L. Lin and M. A. Goodrich. UAV Intelligent Path Planning for Wilderness Search and Rescue, Proceedings of IEEE/RSJ International Conference on Intelligent Robots and Systems. Oct, 2009 St. Louis, Missouri, USA.

L. Lin and M. A. Goodrich. A Bayesian Approach to Modeling Lost Person Behaviors Based on Terrain Features in Wilderness Search and Rescue. Proceedings of the 18th Conference on Behavior Representation in Modeling and Simulation. Sundance, Utah, USA. March 31-April 2, 2009. pp. 49-56.

J. W. Crandall, M. A. Goodrich, and L. Lin. Encoding Intelligent Agents for Uncertain, Unknown, and Dynamic Tasks: From Programming to Interactive Artificial Learning. In Proceedings of AAAI Spring Symposium: Agents that Learn from Human Teachers. March, 2009. Stanford, California, USA.

L. Lin UAV Intelligent Path Planning for Wilderness Search and Rescue. Master Thesis, Brigham Young University, 2009, Provo, Utah, USA.

Personal tools