As the development of autonomous driving technologies became a trend in recent years, the competitiveness of sensor components that collects information of driving environment then conduct the fusion of data has been increased continuously due to the efforts of IT industry. The sensor components can be divided into four types: cameras, LiDARs, radars, and ultrasonics. Except cameras, the other three categories utilized the theory of emitting signals to collide objects then reflect it for receiving so as to measure distances and speeds of the objects. LiDARs, the most expensive one, further establishes 3D models of the environment on this basis, to categorize objects in more details, providing AI brains to read and decide precisely. LiDARs nowadays can be divided into two types: mechanical and solid-state, and the latter doesn’t need maintenance and it’s cheaper due to the simple design, and it has become the mainstream gradually. As for radar, it had huge leap in technology development in the past two years that its data collection is more complete and precise; for instance, 4D imaging radars, the hot topic recently, can master data comprehensively including distances, orientations, speeds, and even heights. In the resolution of angles of objects can be as precise as 0.1 degree, plus the cost is much lower than LiDARs, so it plays a more critical role in the combination of sensor components on cars gradually. Certainly, the innate disadvantage of radars is that it can be affected by other radar emissions or bad weathers so that the accuracy of data will decrease significantly. By the efforts of R&D personnel around the world, they not only using the application of CNN (Convolutional Neural Networks) of AI to learn the processing of noises, but also thinking “out-of-the-box” in new direction of development - quantum entanglement. This theory that was proposed by Albert Einstein in 1935 is now utilized to emit entangled photons to improve the accuracy of data collected by radars. If this research is successful, then the accuracy will be theoretically 500 times higher. These development in progress is predicted to have achievement at least 3 years later, meanwhile the commercialization of autonomous driving technologies more likely takes double the time, so the advancement of autonomous driving technologies will go faster due to these improvements of the hardware technologies.
At the moment that the hardware technology of sensor components keeps improving, the fusion computing of data received from various components also keep developing so as to increase the accuracy and reliability of the cognition of the whole driving environment of autonomous driving systems, then make real-time, correct, and safe decisions. To be simple, data fusion is one of the most critical technologies to determine the timetable of the commercialization of autonomous driving technology in the future. The challenge of technology comes from…
- Various types of sensor components have different frequency of data receiving.
- Data received by sensor components might be wrong or even unreceived.
- Objects are in the blind zone of sensor components and move constantly.
- The quality of data receiving isn’t consistent under different environment.
- Sensor components will have worse quality of data receiving due to durability.
Because there are many uncertainties in existence, the design of fusion algorithms is therefore very complicated to be taken care of; and the levels of autonomous driving technologies from various companies can be identified from here. Although the challenge is so hard, experts will find the right way in the end as verification and adjustments continue on and on in the simulated road tests and real vehicle road tests; It depends on how decision-makers in companies get through the pressure challenge of the prolonged timeline and massive cost of development anyway.