In the Three-dimensional World, the Handwork of the Robotic Arm Is Also Invincible

By Super Neuro

Many people must have experienced the block-building game Jenga, because if you are not careful, the block tower you have worked so hard to build will be destroyed. What will happen if AI and robots are given the task?

It seems that teams studying artificial intelligence always like to find breakthroughs through games. The robotic arm developed by the MIT team also started from games in the three-dimensional world.

In stacking games, you usually stack the blocks three at a time to form a tower, then take the blocks from the bottom and put them on top of the tower to create a taller tower.

The game of Jenga is a test of patience, balance, strength, and many other aspects. For many people (especially those with trembling hands), this game is too difficult. However, this robot developed by MIT easily overcomes this task through detection, algorithms, push-pull, alignment, and other operations.

Who exactly is it?

Humans always say that they have "shaky hands", so the research on robotic arms is to complete some delicate or high-risk operations. Alberto Rodriguez, an assistant professor of mechanical engineering at MIT and one of the project team members, pointed out that the key to this robot is that it perfectly combines vision and touch.

But from the appearance, this robot is similar to some common application machines, like an ordinary robotic arm, but it is equipped with a soft-toothed gripper, a force-sensing wristband and an external camera, which is equivalent to giving it hands, touch and eyes.

When working, the gripper is used to manipulate the blocks and also provide tactile feedback; the sensor wristband is used to control the force of manipulating the blocks; and the camera is used to collect visual images.

In addition to its appearance that allows the robot to move blocks flexibly, the most important thing is that it has a "soul" that is different from previous robots - researchers use new algorithms to make it better at this job.

According to MIT researchers, this robot does not use traditional AI learning methods, but creatively uses hierarchical model dynamics to build a cluster learning model. The advantage of this is that it no longer relies on a large amount of data, but can make real-time analysis based on feedback data, and predict the solution to move the next block while detecting contact.

How does it play Jenga?

In fact, the robot can handle the seemingly complex Jenga game, and the key is the use of cluster learning.

The traditional way to solve this game is to collect all the relationships between blocks, robots, and block towers, and then calculate the best way. But this will obviously bring a huge amount of data and greatly increase the difficulty of calculation.

In this study, the robot was chosen to imitate the way humans play games. First, the data was labeled and clustered through trial and error. Then the feasibility of the new operation was judged by comparing it with the labeled data.

First, let the robot face a building block tower, randomly select a building block and push it out with a relatively small force. For each pushing and pulling operation, the computer will record the corresponding visual and force data, and mark it together with the result of the operation.

This study took about 300 attempts to accumulate enough data, and then the data was processed. Clustering was used here. The data and operations with similar results were grouped together to represent specific building block behaviors.

Different groups represent different degrees of maneuverability, which is also the standard for measuring each operation. For example, one set of data represents the robot's attempts on a difficult-to-move block, while another set of data represents an attempt on an easier-to-move block.

For each different data set, a simple model is given accordingly. By combining these models, the robot is able to learn in real time.

Finally, you can conduct actual drills. When the robotic arm pushes out the building blocks, it uses the camera and wristband to receive visual and tactile information, and then compares the received feedback with the previous data. If the data corresponds to a good result, the operation is performed. If there is a risk of collapse, the operation is abandoned.

It’s not just about Jenga

MIT researchers pointed out that although the robot was able to play this game in the study, it would still need some improvements if it were to compete with human masters. Because in this study, the AI robot focused on solving the problem of physical interaction, solving problems such as whether the building blocks can be pulled out and put up. But the game of Jenga still requires some strategies, which involves considering and analyzing related steps.

But the MIT research team obviously didn't have this idea. Perhaps for them, creating a master of playing Jenga is not of much value. According to Rodríguez, a researcher on the team, they are considering using this technology in actual work environments, such as in the fields of robots in manufacturing assembly lines.

Wait a minute, it's just a simple puzzle game, we may not be able to beat it, it won't play with us humans. Seeing it is about to do something more amazing, Emmmm, forget it, let us be overwhelmed by the New Year's food.