A Perspective on Safety and Real-Time Issues for GPU Accelerated ADAS

Abstract

The current trend in designing Advanced Driving Assistance System (ADAS) is to enhance their computing power by using modern multi/many core accelerators. For many critical applications such as pedestrian detection, line following, and path planning the Graphic Processing Unit (GPU) is the most popular choice for obtaining orders of magnitude increases in performance at modest power consumption. This is made possible by exploiting the general purpose nature of today's GPUs, as such devices are known to express unprecedented performance per watt on generic embarrassingly parallel workloads (as opposed of just graphical rendering, as GPUs where only designed to sustain in previous generations). In this work, we explore novel challenges that system engineers have to face in terms of realtime constraints and functional safety when the GPU is the chosen accelerator. More specifically, we investigate how much of the adopted safety standards currently applied for traditional platforms can be translated to a GPU accelerated platform used in critical scenarios.

Contending memory in heterogeneous SoCs: Evolution in NVIDIA Tegra embedded platforms

“…In latency-sensitive scenarios such as Advanced Driver-Assistance Systems (ADAS), avionics and industrial automa-tion, it is crucial to precisely estimate the execution time requirements of each task on a target computational platform [8]. This requirement stems from the need for an accurate Worst Case Execution Time (WCET) estimation that could inform the decision of job schedulers able to guarantee hard real-time requirements [13].…”

“…For instance, more than one CPU core sharing a common cache level causes uncontrolled eviction of useful cache lines; at system memory level, undisclosed arbitration policies in memory controllers might severely impact latencies when multiple clients are accessing memory in overlapping time windows. A platform-specific characterization of such problems is mandatory before attempting to design adhoc mitigation solutions for memory contention [8], [9].…”

GPU Devices for Safety-Critical Systems: A Survey

“…High-performance computing devices such as GPUs and multi-core devices are increasingly considered for the development of safety-critical systems in multiple domains such as transportation (e.g., automotive [15,203,141,13], railway [20], avionics [21,27], space [105,196,9]) and industrial machinery (e.g., wind turbines, industrial control robots) [153,11]. This trend answers the need for higher computational performance and higher on-chip integration of functions required to facilitate the development of next-generation safety-critical systems.…”

“…This trend answers the need for higher computational performance and higher on-chip integration of functions required to facilitate the development of next-generation safety-critical systems. For example, the automotive domain Autonomous Driving (AD) and Advanced Driver-Assistance Systems (ADAS) require unprecedented levels of computing performance to process computationally-demanding algorithms such as computer vision perception and Machine Learning (ML) algorithms [121,141]. Moreover, there is also a cross-domain trend towards the integration of different safety criticality functions (a.k.a., mixed-criticality) in a reduced number of high-performance computing devices reducing hardware's solution size, weight, and power costs and potentially increasing the overall system reliability by a reduction of components, cables and connectors [153,11].…”

Reconciling QoS and Concurrency in NVIDIA GPUs via Warp-Level Scheduling

“…Modern automotive applications feature computeintensive workloads in which tasks must be processed within defined timing requirements. Applications such as object tracking, lane-following, and obstacle avoidance are safety-critical and must therefore meet hard deadlines [1]. In contrast, augmented/virtual-reality applications like rendering, SLAM and eye-tracking feature soft deadlines or softer quality-of-service (QoS) requirements.…”