TU-TESL

• 22

According to foreign media reports, Fraunhofer Institute for integrated circuits IIS (hereinafter referred to as Fraunhofer IIS) is jointly developing a software programmable and reconfigurable hardware platform with partners of ki-flex project, using AI method to process sensor data, helping to measure vehicle position and determine vehicle environment, ensuring safety and reliability. (picture source: newmobility) automatic driving requires fast and reliable processing and fusion of the data of vehicle laser sensor, camera and radar sensor, so that the vehicle can continuously obtain accurate images of actual traffic conditions, so as to determine its position in the environment and make correct decisions under different driving conditions. Dealing with these data is very complex and requires an AI approach to ensure road safety. For this reason, ki-flex project is committed to developing a powerful hardware platform and related software framework. Its algorithm for sensor signal processing and sensor data fusion is mainly based on neural network, which can determine the exact location and environment of the vehicle. The relevance and availability of individual sensors depends on traffic conditions, weather and light conditions. For this reason, the platform is designed as software programmable and reconfigurable hardware, and its sensor evaluation algorithm can switch according to the changing driving conditions, so that the vehicle can still respond flexibly when a single sensor is damaged or fails. In addition, the project team will develop appropriate methods and tools to ensure the functional safety of the AI algorithm and its interaction, even if the algorithm is reorganized during vehicle driving. The computing resources of the hardware platform will be dynamically allocated according to the load, so as to effectively perform all algorithms and reorganization. This platform is a new development in the field of neural morphological hardware. Its function is inspired by human brain, and it is specially designed and optimized to effectively use neural network. The project also considers that although the product cycle of automobile industry is long, AI algorithm is developing rapidly. As a result, project partners are committed to developing hardware platforms that can quickly and easily adapt to new software and hardware requirements in the field of machine learning. For this reason, researchers use a flexible programmable multi-core deep learning accelerator, which is in the form of a specially developed chip (ASIC). Compared with the traditional multi-purpose processor (CPU) or graphics processing unit (GPU), this ASIC can reduce the cost and power consumption. All in all, the project plays an important role in promoting the development of autonomous driving.