Marvel Digital AI Limited
RDA TECHNOLOGIES LIMITED (CS)
As the society is ageing rapidly, there is an expansion of the manpower resources gap in hospitals, elderly nursing homes and etc. It leads to a definite market demand on smart AGV (Automated Guided Vehicle) to take over human manual labour works such as transportation, sanitizing, cleaning, delivering and so on. Only in mainland China, the market capacity is expected to be many hundred billion Hong Kong dollars. The application scenarios will require a more intelligent perception system for the smart AGV to accomplish missions in complex environments such as in hospitals. However, the industry is lack of dedicated computation platforms to support those intelligent functionalities and make the mass production cost-effectiveness. This project aims to develop a low-cost hardware-enabled platform (verify on FPGA platform) for smart AGV perception module targeted for low-speed < 4km/h indoor AGV applications such as obstacle detection, avoidance and indoor navigation inside offices or hospitals. We leverage and optimize the state-of-the-art perception algorithms of multiple sensors including LIDAR and camera for indoor usage scenarios, and then propose a dedicated perception accelerator platform to handle the computations efficiently. We develop a template-based complier supporting ONNX and BLAS, system software including APIs, drivers, and programming scripts, and a machine learning system to embed the perception accelerator into the mainstream application eco-system seamlessly. The project outcomes relieve pain points at the computation platform and perception technology in smart AGVs. This project provides a practical solution for smart AGV IC and perception module manufacturers worldwide and applicable in both Hong Kong and mainland China. It also helps to cultivate high-tech innovations in both smart city, smart manufacturing, and semiconductor areas in Hong Kong and GBA.