Once humans mastered the tricks when they were toddlers, they became very good at the "motion planning" that robotics said in their mouths: groping around obstacles and accurately catching them in a refrigerator filled with things. Take a bottle of drink, or reach into the back of the computer screen and touch the cable outlet and connect the cables.
But for robots with multi-joint robotic arms, motion planning is very difficult and requires very time-consuming calculations. In an environment that has not been specifically optimized for robots, it takes even a few seconds for the robot to pick up an item.
Researchers at Duke University have developed a computational processor specifically designed for motion planning that is 10,000 times faster than existing devices and consumes much less power. This new computing processor is very fast and can be planned and executed in real time. And its power consumption is so low that it can be used in large-scale manufacturing environments that include thousands of robots.
“When you're thinking about a car assembly line, the entire environment is tightly controlled so that the robot can repeat the same action over and over again,†said George Konidaris, assistant professor of computer science and electronics and computer engineering at Duke University.
“The parts of the car are in the same place every time, and the robot is confined to the cage so that it won’t be disturbed by others. But if your robot performs real-time motion planning, whether the auto parts are in the same place, Unexpectedly piled up casually, or someone walks by, it always makes the right move."
According to Konidaris, rapid motion planning saves the time and cost of deploying the environment around the robot. He was held in Ann Arbor, Michigan June 20 showed his new study:: (Science and Systems Robotics) conference on "Robotics Science and Systems Engineering."
Research in the field of motion planning has been going on for 30 years, and recent advances in this field have been able to reduce the planning time of complex robots to a few seconds. In addition to a few exceptions, these existing methods typically rely on a general-purpose CPU, or a faster, but more energy-intensive graphics processor (GPU).
Duke’s team has created a new processor specifically designed for sports planning.
"Universal CPUs are good at handling multiple tasks, but they are not comparable to processors that are good at single tasks," said Daniel Sorin, a professor of electrical and computer engineering and computer science at Duke University.
The team of Konidaris and Sorin allowed the processor to perform the task of collision detection, which is the most time-consuming part of motion planning. The processor is capable of executing thousands of collision detection tasks in parallel.
Sorin said, "We have optimized the design so that the hardware and power budgets are dedicated to these specific tasks related to motion planning."
The principle of this technique is such that the operating space of the robot arm is divided into thousands of 3D spaces called voxels. The algorithm then determines if an object is in the voxel covered by the preprogrammed motion path. Thanks to specially crafted hardware, the technology is capable of detecting thousands of motion paths simultaneously and then integrating the shortest possible motion paths with the remaining "safe" options.
“Previously the most advanced technology used high-performance commercially available graphics processors, which consume 200 to 300 watts of power,†Konidaris said. “Even so, it takes hundreds of milliseconds, or even seconds, to find a sport. Planning solution. We designed the processor in less than 1 millisecond and consume less than 10 watts. Even if our speed is not faster than theirs, the light energy saving can give factories with thousands or even millions of robots. Save considerable costs."
Konidaris went on to say that the technology opens up new ideas for the application of motion planning.
“Previously, motion planning was done separately for each action because the planning process was very slow,†Konidaris said. “But now it's fast enough to be part of a more complex planning algorithm that might be able to Several simple actions are concatenated, or you can pre-infer the planning of the actions of several objects."
The speed and energy efficiency of this new processor brings many opportunities to the automation field. Konidaris, Sorin and their students are very optimistic about the technology and have set up a company called RealtimeRobotics for this technology.
Half-cut Cells Mono Solar Panel
Half-cut Cells Mono Solar Panel
Half-cut cells mono solar panel
Zhejiang G&P New Energy Technology Co.,Ltd , https://www.solarpanelgp.com