Speaker

Dr. Liangkai Liu, Research Fellow, Department of Computer Science and Engineering, University of Michigan

Abstract

Deep Neural Networks (DNNs) are extensively used in autonomous vehicles (AVs) and autonomous mobile robots (AMRs) for environmental perception. However, reducing the end-to-end (e2e) latency for multi-camera-based Bird's Eye View (BEV) perception in AVs, while maintaining accuracy, remains challenging. In the first part of my talk, I will introduce RT-BEV—the first framework that co-optimizes message communication and object detection to enhance real-time e2e BEV perception without compromising accuracy. For battery-powered AMRs, improving energy-efficiency for autonomous navigation while preserving accuracy presents similar challenges. In the second part of my talk, I will present Donkey, which integrates milli-second level energy prediction into the path planning to improve efficiency.

Bio

Liangkai Liu is currently a Research Fellow in the Department of Computer Science and Engineering at the University of Michigan, working with Prof. Kang Shin. He earned his Ph.D. from Wayne State University under the guidance of Prof. Weisong Shi. His research centers on building safe, predictable, and efficient autonomous cyber-physical systems. Notably, he is among the first in the computing systems community to explore the issue of DNN inference time variations in autonomous driving, leading to publications in RTSS, RTAS, and ICCAD, as well as securing an NSF grant. His work on energy-efficient autonomous systems has been featured in ACM SEC, ICRA, TITS, and HotEdge. He developed testbeds like HydraOne and Donkey, which serve as programmable and energy-efficient platforms for autonomous mobile robots. Learn more about him by visiting his website.