Research

My current research interests are robotic search, human search behavior analysis and reinforcement learning.

When I was a first year PhD student, I tried to enable a mobile robot with Kinect sensor to search for a target.

However, after several months, I realized finding an optimal search is extremely difficult.

At the same time, I have a one year old daughter. We play a hide-and-seek game everyday.

One day, I found she can catch me within 15 seconds no matter where I am in my apartment.

It's amazing! She is faster than my robot!

Then, I started to think about how human can search for a target so fast and how to improve my robot's search algorithms.

"Search" becomes my PhD thesis, which is consisted of robotic search and human search.

[Robotic search]

The goal of robotic search is to enable a mobile robot to search for a target as soon as possible.

It includes coverage problems, probabilistic search and minimum-time trajectory planning.

It is reformulated as a unified objective function.

The robot learns PD and CTG functions and search for the target based on learned functions (see the video).

[Human search]

The goal of human search is to explore how humans search for a target so efficiently.

10 human subjects are recruited and search for the target via teleoperation.

Human gaze, control, robot position and RGBD image are stored for analysis.

According to the results, subgoal hypothesis is proved and a new objective function for search is found.

Based on the objective function, near-optimal subgoals are generated for helping human search.

The results demonstrate that human performance is improve 4~40% with subgoal assistance.

My previous research (2002-2011) is here.