Robot grab

In the past two days, we visited Huawei’s automated factory and learned about the application of robotic grabbing in the industry and the prospects for the future. We are encouraged. As an old “grasper” who has been robotically grabbing for nearly eight years, I think it is still possible to write robots to capture past and present lives. I also hope that more friends who are interested in this matter do not think robots have no future in their grasp. The road ahead is still long! (Grasping is not solved! Didn’t know if you knew anything about this history:)).

After Stanford made the first robot arm for pick-and-place around 1980 (forget the name), only a two-gripper was used, as if only one was grabbed and stacked together. Video. At the same time, MIT Salisbury and the people hunted down the JPL hand. Like the picture below, three fingers can turn the cuboid block through the hacking code.

However, hacking hacking (the word really does not know how to translate properly), or someone has extracted a scientific problem:Given a known or unknown object, how to select the optimal grabbing point and apply the appropriate force so that the object can achieve the desired behavior?(See below). These expected behaviors include maintaining stability, doing some actions, or tasks.

Since then, Grasping has started a 30-year irrigation career. Every time ICRA, IROS (Robots Mainstream Conference), there is a separate session about grabing. At the beginning, one of the main directions in this field is to study the mathematical form of this problem, including various grabbing constraints and objective functions, covering domain-containing analysis, planning and control, as shown in the figure below. Until Professor Li Zexiang wrote this book, Dacheng’s robot-based mathematics-based teaching material, this kind of research is slowly becoming less hot. In this direction I generally call them mathematical analysis schools and I am still very active. Do not know why, many Chinese students are in this direction.

From another aspect, many people have also done hardware design of various robotic hands, trying to end the problem from the hardware. This direction began with salisbury at MIT and later Barrett hand. Later I slowly studied the transfer of the town to Europe. The DLR teacher Liu Hong made a smarter DLR hand. The current version is also getting better and better. Of course, the Japanese also studied several dexterous hands at the same time, including Gifu hand. One of the core features of these hands is to be tall and dexterous, at no cost, and the price of each hand is therefore very high, much higher than the price of the robot arm. The price of Gifu hand is almost one million. Another feature is that due to the large amount of complex integration in a small space, the entire system is usually not very robust and needs to be used with care.

Hardware Improvements With Yale’s creativity and the transformation of the University of Pisa recently, a variety of cheap robots began to appear. Here is a free 3D printer, but Italians sold it for 60,000 yuan. This should be placed in the country, that is, a toy.

The hardware upgrade brings a variety of hand-held sensor improvements, and the more specific hand is the tactile sensor. Touch sensors at the ends of various fingers and tactile sensors on the entire palm (such as TexScan) have also been produced. These sensors are still very expensive in Japan, but one difficulty is that the model of the sensor is difficult to establish, or that the tactile sensor is not very easy to use. In the end, how to use tactile information has not yet been solved.

Speaking of so many research directions of robotic grabbing, in the industry, most of the grabbing operations have been solved by special equipment. It is necessary to transport belts and other sorting devices. Even with gripping, two-finger clamps or suction cups are usually used. So I had the first paragraph of IROS classic quotes: Grasping is solved! This fact actually makes people who are doing grabbing a bit embarrassed, because usually people think that this study has no practical value. It’s not that the hand has converged in the past, and then the object is gone.


We need robots, machine learning and collaborative robot design jobs, friends, friends, welcome to contact!