Tsang-Kai Chang

E-mail: tsangkaichang[at]ucla[dot]edu

Tsang-Kai received his B.S. and M.S. degrees in Electrical Engineering from National Taiwan University in 2012 and 2014, respectively. Currently, Tsang-Kai is a Ph.D. student in UCLA.

His research interest includes multirobot system, stochastic modeling.


General SLAM

Deep neural networks have succeeded in many applications by replacing model-dependent approaches with end-to-end computation architectures. However, the application of deep neural networks on robotics remains challenging. One important task application in robotics is SLAM, which enables robots to acquire the spatial knowledge of the environment automatically. At the same time, the sense of space is extensively studied in neuroscience as the key component in brain-mind interaction.

Even though the approaches are different, the key problems toward general SLAM include:

  • the general map representation,

  • the end-to-end computational method to build the map

General Map Representation

The first main challenge is that there is no unified spatial representation, also known as map. In biological brain, different cells that fires with various spatial clues are observed, including place cells and grid cells. Those cells are also observed in the artificial neural networks in foraging tasks. The existence of those cells suggests the basic representation of the space.

End-to-End Computational Architecture

The other difficulty of extending current deep neural network techniques to solve SLAM-related problems lies in the lack of memory in deep neural network. In other words, deep neural networks basically reflexively output the results based on the input. There are several works to add sophisticated memory elements on neural network to solve navigation problems. However, the results are still at a very early stage. Also, the biological neural systems accomplish computation and memory simultaneously.


Journal publications

  • T.-K. Chang and A. Mehta, “Control-Theoretical and Topological Analysis of Covariance Intersection based Distributed Kalman Filter,” IEEE Control Systems Letters, 2018. [IEEE Xplore]

  • T.-K. Chang and A. Mehta, “Optimal Scheduling for Resource-Constrained Multirobot Cooperative Localization,” IEEE Robotics and Automation Letters, vol. 3, no. 3, pp. 1552 - 1559, July 2018. [IEEE Xplore]

Conference proceedings

  • T.-K. Chang, S. Chen, and A. Mehta, “Multirobot Cooperative Localization Algorithm with Explicit Communication and its Topology Analysis,” in 2017 International Symposium on Robotic and Research (ISRR), Dec. 2017.

  • T.-K. Chang, K.-C. Chen and L. Zheng, "Time Dynamics of Random Access in Cognitive Radio Networks," in Proceedings of 2014 International Conference on Communications (ICC), June 2014. [IEEE Xplore]

Blog posts