Disney Research Teaches Robots to Fall: How It Protects Expensive Mechanisms.

Disney Research Teaches Robots to Fall: How It Protects Expensive Mechanisms
Disney Research Teaches Robots to Fall: How It Protects Expensive Mechanisms

According to Korrespondent.net: Bipedal robots are incredibly complex mechanisms, yet their vulnerability to falls raises many questions. A minor push or obstacle can lead to a fall, often resulting in damage to expensive components such as cameras. According to , to address this issue, researchers at Disney Research in Zurich developed a system that teaches robots how to fall softly and in a controlled manner.

Effectiveness of New Methods

Current methods to protect robots during falls are not sufficiently effective. Some may lead to drive blockages, causing the robot to become a 'statue' and fall. Other approaches rely on pre-defined movements that may only work in simple situations.

In the new study, the Disney team employed reinforcement learning methods. In a virtual environment, thousands of digital robots fell from various positions and angles. Each of them received a 'reward' for reducing impact and landing successfully in one of the designated artistic poses. Through a constant cycle of trial and error, the robots learned controlled falling strategies in various conditions.

Testing on Real Robots

The developed strategy was transferred to a real bipedal robot, and a series of tests were conducted. Engineers selected ten aesthetic final poses created by artists and then pushed the robot, causing it to fall from different directions. The robot endured all tests without damage, successfully executing controlled movements during the fall and ending its journey in the desired pose.

'This is the first universal approach that allows controlling the fall of a bipedal robot in real conditions,' the authors note.

The next step for the team is to test this method on other types of robots, including quadrupeds, to assess its universality. Researchers are also working to teach robots to anticipate falls, as currently the system activates only after the loss of balance has begun. A separate goal is to create an algorithm that enables the machine to gracefully rise after a fall.

It is worth noting that earlier reports mentioned a robot from South Korea that amazed everyone with its 'moonwalking' in a video.

Implementing such technologies could significantly change approaches to the use of robots in various fields, including service and industry. Utilizing controlled falls will help reduce risks to expensive components, which in turn will lower maintenance and service costs. This opens new perspectives for the commercial use of robots in everyday life.


Read also

Advertising