Compared with CPU and GPU, FPGA has obvious advantages in simple and repeated tasks such as perception processing. According to the current trend, FPGA may replace GPU in robot development in the future. Although both FPGA and GPU are good at a large number of repetitive operations, the energy consumption of FPGA will be much lower than that of GPU. However, FPGA may never be able to replace CPU. In fact, in various applications, FPGA also appears as a coprocessor of CPU rather than a real core computing unit. The reason is actually very simple. It is a word we mentioned several times before: design purpose.
The ultimate goal of designing robots is to hope that they can become like us, think in a similar way, and have the ability to independently judge the situation and deal with tasks. At present, we don’t know the specific principle of the brain, but whether from experience or intuition, human like thinking is not simply piled up by countless logic. Because the complexity of thinking has gone far beyond the scope of operation.
For example, if you want to calculate 213 x 312 and you want to simplify it, you just need to decompose it into 213 312S and add them, or even repeat 213 times and 312 ones. But what if you want to calculate the 64th power of 2? What if you want to calculate sin (27 °)? Even if you want to calculate log230? If you must divide it into simple operations, you will find that the resources you need will increase at an unimaginable rate until you can’t even call the primary school students in the city.
According to the current performance, the human brain is obviously not suitable for large-scale parallel computing. It is more like a CPU that is good at single and deep thinking (operation). And the function of human brain is more consistent with the design and positioning of CPU. Therefore, in any case, before there is a breakthrough in brain science, people will still tend to use the CPU that seems to have more potential as the core computing component of the robot. Perhaps with the development of FPGA, the data operation of more and more sensing components of robots will be undertaken by FPGA (just like human cerebellum), but it is obvious that the status of a component is mainly and core or not, which is not judged by quantity.
It cannot be denied that FPGA can indeed play a great role in robot development. But on the whole, it is more like a position of performing its own duties with CPU and even GPU. It may be inaccurate to talk about substitution at any time.