FAST AND LOW-POWER DEEP LEARNING SYSTEM ON EMBEDDED HARDWARE FOR SELF-DRIVING AUTONOMOUS BICYCLE

Yucheng Yang, Kiran George, Kenneth John Faller II, Pradeep Nair
2022 Zenodo  
Tesla, Google, and Waymo are all attempting to develop self-driving cars that can navigate real-world roadways. Many analysts anticipate that fully driverless cars will be on the road in our cities within the next five years and that practically all automobiles will be autonomous within 30 years. Automatic driving is a massive and complicated endeavor that incorporates various technology. Environment perception, behavior judgment, path planning, and motion control are the four essential
more » ... d driving technologies. Collecting and analyzing environmental and in-car data is the initial step in environmental perception, which is the foundation and premise of autonomous driving in intelligent vehicles. The optimization of image processing and the selection of hardware technologies are the key topics of this thesis. As a result, a self-driving bike is created to demonstrate the proposed software and hardware co-designed machine learning. Moreover, the proposed soft and hardware co-designed machine learning model is implemented on a development board to obtain greater energy savings and more precise data processing. To achieve this, we also propose a new implementation based on existing Xilinx software to minimize the development cost, which differs from typical FPGA transplanting technology.
doi:10.5281/zenodo.6950722 fatcat:xpz3p7x6s5dszmcohy4zoetfhu