While generative artificial intelligence (AI) is advancing rapidly, physical AI is still lagging behind. Current AI technology struggles to predict and make judgments in complex work environments. This is particularly true in disaster sites or outdoor workplaces, where even a small misjudgment can lead to significant accidents, hindering commercialization.
As a result, semi-autonomous robots, which allow humans to directly control the machines while AI assists with certain tasks, are gaining attention as a practical alternative.
According to industry sources, U.S. robotics startup Minerva recently developed a humanoid robot named 'Roger.' Unlike fully autonomous robots, Roger does not independently assess its surroundings. Instead, it operates by mimicking the movements of a skilled worker wearing virtual reality (VR) equipment.
When the operator raises an arm, Roger raises its arm as well; when the operator walks, Roger moves accordingly. The worker can monitor the site through a camera mounted on Roger and control the robot's movements with their own body. Human judgment and gestures are transmitted to the robot, while the AI onboard manages balance, movement, and posture recovery if the robot falls. This setup combines the 'brain' of a person with the 'body' of a robot.
Roger is equipped with five fingers, enabling it to perform tasks designed for human hands, such as turning valves or pressing buttons. Minerva explains that Roger's dexterity allows it to be deployed in bomb disposal, offshore oil drilling facilities, oil and gas operations, and hazardous material response scenarios.
Currently, robots with four legs are primarily used in dangerous environments. While they excel in mobility and surveillance, they have limitations when it comes to grasping tools or manipulating equipment. Roger's humanoid form and hands differentiate it by being able to replicate human work methods in areas that are difficult for humans to access.
Minerva's choice to prioritize remote control over full autonomy aims to expedite the deployment of robots in the field. Physical environments often present situations not covered by training data. Slippery floors, sudden obstacles, and equipment that operates unexpectedly can occur. In such cases, it is safer for an experienced person to make judgments rather than relying solely on AI.
The remote control method compensates for the robot's lack of autonomy with human experience, allowing operators to work from a safe distance. This reduces the need for bomb disposal technicians to enter hazardous sites in protective gear or for workers to be exposed to toxic gases and high temperatures. It represents a realistic pathway to replace dangerous tasks with robots before achieving full autonomy.
U.S. robotics company Apptronik is also applying remote control methods in its development of the industrial humanoid 'Apollo.' Operators remotely control Apollo to perform tasks on production lines. Apollo learns to repeat the same tasks autonomously based on the data accumulated during this process.
Remote control capabilities are also being integrated into home humanoids. U.S.-Norwegian robotics company 1X Technologies has developed a home humanoid named 'NEO,' which features a remote control function called 'Expert Mode.' When NEO encounters a task it has not faced before or a situation it cannot resolve independently, a remote operator intervenes to control the robot.
As a result, before fully autonomous robots become a reality, it is likely that physical AI, where humans make decisions and robots execute tasks, will see widespread adoption. A report published in June by the World Economic Forum (WEF) and Accenture, titled 'Human-Machine Collaboration in Industrial Operations,' noted, 'As physical AI is introduced in production environments, robots will be taught behaviors through VR remote control, and the performance of autonomous systems will be monitored. New roles such as 'robot engineer-orchestrators' may emerge to intervene in situations requiring human judgment.'
* This article has been translated by AI.
Copyright ⓒ Aju Press All rights reserved.