I am a P.h.D. candidate in computer science at Georgia Institute of Technology, supervised by Dr. Richard Vuduc. I am now on job market.
I enjoy building systems. I am broadly interested in computer systems and deep learning, such as solving path planning problem, collision detection and system techniques.
Contact: CODA S1275F, xchen384@gatech.edu
Previous path planning algorithms of inspecting 3D objects ignore constraints on the robots. Avoid collision is usually mandatory and demands an expensive cost in the processes of covering 3D objects. We propose a new algorithm that presents an efficient way of partitioning the object and how to reduce the cost of avoiding collisions. [under review]
Most coverage path planning approaches employ hard-coded heuristics or other application-specific requirements, making them hard to extend to other problem scenarios or configurations, such as different motion strategies or robot sizes. This work presents a unifying, general, and adaptive frameworks using deep reinforcement learning, called ADPath, for coverage path planning problems under a variety of configurations. [IROS 19]
Most collision detection methods require three types of geometrical operations that are bottlenecks: decompositions, rotations, and projections. We present a new and more work-efficient parallel method to speed up a class of three-dimensional CD problems. Our approach simplifies these operations and, empirically, can prune as much as 99% of the collision tests that would otherwise be required and improve load balance. [ICPP 19]
Checkpointing in stream processing can cause significant interferences for executions that degrade system performances and even leading to exhaustion of resources or system crash. We propose Governor, a controller that factors the checkpointing costs into the backpressure mechanism. It not only guarantees a smooth execution of the stream processing but also reduces the throughput loss. [ICAC 17]
Xin Chen, Thomas M. Tucker, Thomas R. Kurfess, Richard W. Vuduc, Liting Hu, “Max orientation coverage: efficient path planning to avoid collisions in the CNC milling of 3D objects”, under review
Xin Chen, Thomas M. Tucker, Thomas R. Kurfess, Richard W. Vuduc, “Adaptive Deep Path: efficient coverage of a known environment under various configurations”, IROS 2019 [pdf]
Xin Chen, Dmytro Konobrytskyi, Thomas M. Tucker, Thomas R. Kurfess, Richard W. Vuduc, “Faster parallel collision detection at high resolution for CNC milling applications”, ICPP 2019 [pdf]
Xin Chen, Ymir Vigfusson, Douglas M. Blough, Fang Zheng, Kun-Lung Wu, Liting Hu, “GOVERNOR: Smoother Stream Processing Through Smarter Backpressure”, ICAC 2017. [pdf]
Xin Chen, Liting Hu, Douglas M. Blough, Michael A. Kozuch, Matthew Wolf, “RBAY: A Scalable and Extensible Information Plane for Federating Distributed Datacenter Resources”. ICDCS, 2017. [pdf]
Xin Chen, Ymir Vigfusson, Douglas M. Blough, Fang Zheng, Kun-Lung Wu, Liting Hu, “Smoother Stream Processing Through Smarter Back Pressure”(Poster). OSDI 2016.
Wim Lavrijsen, Costin Iancu, Wibe de Jong, Xin Chen, Karsten Schwan, “Exploiting Variability for Energy Optimization of Load Balanced Parallel Programs”, EuroSys, 2016. [pdf]
Liting Hu, Karsten Schwan, Hrishikesh Amur, Xin Chen, “Elf: Efficient Lightweight Fast Stream Processing at Scale”, USENIX ATC, 2014. [pdf]