Matt Min-Hao Wu

California Institute of Technology
Computer Science, MC 256-80
Pasadena, CA 91125

Biography

I am a second year computer science Ph.D. student at the California Institute of Technology. My current field of research is virtual environment for simulating interdiction of malicious materials. My advisors are K. Mani Chandy in the Computer Science (CS) department and Richard Murray in Control Dynamics System (CDS).

I received my B.S. in computer science from the University of Texas - Austin. During my undergrad, I researched simultaneous localization and mapping (SLAM) techniques for robotics. I have also interned at National Instruments, Microsoft, and IBM during each summer.

Research: Radiation Detection

I have been working on the Radiation Detection project since I came to Caltech at 2007. I am the main programmer on the virtual environment for the detection project, using the game engine of Half-Life 2.

So far we have been able to realistically simulate radiation sources' interaction with radiation detectors. This includes:

Modelling radiation as a poisson process, with intensity received on the detector proportional to inverse distance squared

Modelling different shapes of radiation detectors:

Spherical detector- receives the same intensity regardless of orientation
Unshielded panel- receives intensity based on the facing of the panel
Shielded panel- receives radiation only on the unshielded side

Radiation absorption simulation: intensity received by detector is exponentially reduced by the depth and density of materials to the source

Radiation spectrum generation: when simulating photon hits, the detector can record the energy of the photon, generated based on the radiation spectrum of the source material.

Background radiation: the intensity and energy of background radiation can be determined by ray tracing at materials around the detector

The simulation environment also provides visualization to where the radiation source could be based on the hit/miss history of a detector. This is calculated by Bayesian updates to a probablistic heatmap. We have implemented two kinds of heatmap. A grid heatmap and a hierarchical heatmap. The grid heatmap is a fixed-sized grid of points that store the probability of the source being in each location. The hierarchical heatmap is similar to the grid heatmap, but it is only finely divided at important locations. This speeds up computation speed. See videos linked below to see demonstrations of the heatmaps.

Finally, we have implemented search algorithms to pin-point the location of the radiation source. One of the algorithm is a simple gradient descent algorithm, but it is too short sighted to get effective results. A better algorithm is entropy reduction on the heatmap. Since each point on the heatmap represent a probability of the source present, we can coordinate mobile detectors to move in the direction that will reduce the entropy of the heatmap maximally.

Demo Video 1
Demo Video 2