Advancements and Limitations in Sensor Fusion for AI-Controlled Vehicles

Abstract

People’s dreams of self-driving cars navigating the streets of their city are getting closer and closer every year. Regardless, we are still far from reaching that goal due to some hurdles that Artificial Intelligence (AI) needs to clear like visibility and detection of surroundings. This paper gathers and summarizes the advancements in sensor technology and data fusion that contribute to the safety, reliability, and limitations of self-driving cars. Sensor technology is the core for self-driving, as it uses multiple sources to collect information so that the cars do not present a danger to others. As of now, the perception stage of the automobiles is still under development. Data fusion is essential to self-driving, as it combines data from the outside world and is able to translate it into information to which the AI can interpret and respond. Recently, newer, compact cameras have been able to capture more information with better resolution. Although this is a positive for self-driving, cameras still have their limitations. Natural obstacles and harsh conditions (frequent problems on the road) affect the camera’s visibility and detection capacities. Along with having to battle these conditions, long-term exposure to real-world environments cause harm to the camera’s ability to function as intended. I am writing this paper to convey the knowledge we need in order to make this large advancement in the automobile industry. In order for this to be possible, camera technology must become stronger, more reliable, and more affordable. The combination of the three will represent the qualitative change that needs to happen before self-driving cars become the norm.