The speed of some robots like those that make picking of parts is amazing, however the movement to be done is simple. Once a robot must analyze its environment to achieve a complex manipulation, gestures become slow, agonizingly slow. Simply review the images of the DARPA Robotics Challenge to see how this torpor can be sometimes laughable. Researchers propose a new approach to speed up the movements of robots, a solution that relies on a FPGA chip.
The speed of the movements of the robot has been multiplied by a factor of 1 300 to 2 300!
The FPGA chips or programmable logic circuits allow to perform calculations in a way very effective, IE with very high performance for energy expenditure less than that of a conventional microprocessor. the approach in the field of high-performance computing, network equipment and could also interest the roboticists. Indeed, researchers at Duke University, in Durham, North Carolina has just published the results of their work. They wanted to harness the power of FPGA chip to help the robot to understand its environment. Indeed, the slowness of the movements of robots comes to the mass of calculations necessary for the robot to calculate its position and the movement he has to make to make the task expected of him without colliding with its environment. Performed by the means of a FPGA chip, this calculation of path sped 3 orders of magnitude for a reduced power consumption by a factor of 20. The calculation of the motion of a robot arm that must grab an object on a table to place it somewhere else is performed by software on a Xeon 4 hearts to 3.5 GHz in 815,000 microseconds or 756.000 microseconds, according to the chosen algorithm. This same calculation is processed by an FPGA chip in only 622 microseconds. To avoid obstacles, the calculation pass of 2.738.000 / 1.074.000 microseconds to 627 microseconds.
Potentially, it will be possible to do better yet since the installation created by the researchers was a prototype not optimized. The available power can be used to analyze images of higher resolution without increasing the computing time. On the other hand, the device is very flexible. Any change made to the robot requires a reprogramming of the chip. The rest technique still has be improved, but robots will receive a major boost with the arrival of specialized chips.
Translation : Bing Translator
Source : “Robot Motion Planning on a Chip”, Duke Robotics / Duke University research paper