Intelligent cyber-physical systems (CPS)---such as unmanned aerial vehicles (UAVs) and autonomous cars---require high computing performance to execute computationally demanding tasks in real-time (e.g., vision based sense-and-avoid) while reducing size, weight, and power (SWaP) and cost of the system. Although modern embedded computing platforms, which may include multiple CPU cores and GPUs, and FPGAs, can meet the requirements, they also present serious safety challenges for safety-critical CPS due their complexity and unpredictability.

In this talk, we will discuss the safety challenges of intelligent CPS and present two our recent proposals to address the challenges, from the computing infrastructure perspective. The first proposal focuses on the timing predictability of shared caches in modern commercial-off-the-shelf (COTS) multicore architectures. We show that traditional cache space partitioning does not guarantee predictable cache access timing in COTS architectures, and present a hardware (micro-architecture) and software (OS) collaborative solution that improves time predictability at a low hardware cost. The second proposal focuses on the reliability problem, caused by failures in complex software and hardware, of on-board computing platforms in UAVs. We present a low-cost fault-tolerant system architecture using two computing platforms with distinct reliability and performance characteristics. We also present our custom-built flight control system and a fault injection case-study using the system.

Heechul Yun is an assistant professor in the Department of Electrical Engineering and Computer Science at the University of Kansas. He received a Ph.D. degree in Computer Science from the University of Illinois at Urbana-Champaign in 2013. Prior to his Ph.D., he worked at Samsung Electronics and ETRI. His research focuses on OS and architecture support for real-time embedded systems. He is a recipient of the best paper award of the 2016 IEEE Intl. Conference on Real-Time and Embedded Technology and Applications Symposium (RTAS).