2023 Author: Bryan Walter | [email protected]. Last modified: 2023-05-21 22:24
European developers have equipped the Centauro robot with a system that allows the operator to control the robot without additional equipment. It creates a model of a human body in real time and converts its movements into movements of the corresponding parts of the robot. The IEEE Spectrum edition drew attention to the development, which will be presented at the ICARCV 2018 conference.
Engineers are developing not only fully autonomous robots, but also remote-controlled robots, which can have wide capabilities, but their control is completely entrusted to the operator. Their developers see the main area of application for such robots in rescue work or other operations in places where human life is at great risk, but it is necessary to apply rare and complex skills. By themselves, developments in this area already have quite impressive abilities, but controlling such robots is usually very difficult. Some remote-controlled robots obey commands from conventional controllers, and some support exoskeleton control, in which the system monitors all movements of the human body and instantly turns them into robot movements. Despite the fact that this system is considered the most convenient, it requires special training, and is also quite expensive and difficult to implement.
A group of engineers from several European scientific organizations presented the Centauro robotic platform in July 2018. The robot has four legs, each of which has a swivel wheel at the end. In addition, it has a body and two arms, as well as an array of lidar, depth camera and several conventional cameras. Initially, the developers announced the development of an exoskeleton chair for controlling a robot, but now they have shown a control system that mimics the movements of human body parts without the use of controllers.
To recognize movements in the system, the ASUS Xtion PRO depth camera is used, the data from which is transmitted to the OpenPose algorithm, which is often used for such tasks. For example, recently, with its help, a system was created that transfers the movements of one person to a video with another person. The algorithm processes video frames in real time and creates on their basis a two-dimensional model of the body, consisting of many interconnected key points. After that, the system transfers the data of the two-dimensional model to the three-dimensional one and gives this data to the robot's motion planner.
Thanks to this, a person can control the robot without special training, moving his hands and body. The developers showed several examples of movements, for example, a person was able to lift a box and turn a lever. It is worth noting that the OpenPose algorithm allows tracking finger movements, and the Centauro robot has interfaces at the ends of its arms for replacing manipulators, so the system can potentially be used to perform more complex tasks using robotic arms.
In addition to the traditional exoskeleton scheme, which, for example, was chosen by developers from Toyota, Sarcos Robotics and Meltin, there are other unusual approaches to controlling robots. One such system was demonstrated last year by engineers from MIT. They created a virtual cockpit, displayed in a VR helmet, allowing the robot to be controlled with virtual controllers and to see images from multiple cameras in different areas of the cockpit.