The overall objective of the project is to develop techniques and tools to help the system designer in the implementation of safe and secure in-vehicle real-time embedded systems subject to multiple levels of safety certification, in the presence of HW resource constraints, including processor speed, memory size, and energy consumption. The project consists of the following objectives: (1) Objective 1: Efficient Machine Learning/Deep Learning for ADS on embedded systems. In addition to accuracy, it is important to optimize computation efficiency and execution time of Deep Neural Network (DNN) models for deployment on the in-vehicle HW platform for ADS and other purposes. (2) Objective 2: Device-edge-cloud collaborative inference of LLVAMs. The current industry practice of deploying LLVAMs as cloud services suffers from problems of availability, latency, security, and privacy due to network connectivity. Deployment of LLVAMs on edge devices enables local inference without accessing the cloud, which reduces service latency, alleviates performance bottlenecks in cloud servers, and supports local processing of user data for better privacy protection. (3) Objective 3: If necessary for achieving objectives 1 and 2 above, signal enhancement techniques will be designed and implemented on the HW to enhance the desired signal components that can then be used as input to LLVAMs.