Guest Article Why Humanoid Robots Are Finally Within Reach

Related Vendors

NVIDIA GmbH

FAULHABER Drive Systems

Wuxi InfiMotion Technology Co., Ltd.

Eddie Seymour is European Business Director at Nvidia. In his guest article, he explains why humanoid robots are now actually becoming practical thanks to AI, simulation and digital twins.

If you ask people how they imagine a robot, most of them think of a character like C-3PO from Star Wars rather than a production robot in industry or an autonomous vacuum cleaner. Many people still ask themselves the question: why has it taken so long to make humanoid robots usable in practice?

But the wait will soon be over: humanoid robots are moving ever closer to our everyday lives. They are already taking on specific tasks in the first areas—and thanks to the latest advances in artificial intelligence (AI), it shouldn't be long before they are used on a large scale.

New Opportunities Through Humanoid Robots

Our world is tailored to people: Homes, factories, warehouses and transportation systems are based on the human form. If you want to integrate the benefits of automation and technology quickly and smoothly into existing structures, it is therefore best to rely on robotic systems that mimic the human body.

Humanoid robots can help to counteract the growing labor shortage. The World Health Organization estimates that there will be a shortage of around 4 million healthcare and social care professionals in Europe by 2030—including 2.3 million nurses and 1.1 million social workers and assistants. Older people in need of care in particular will feel the consequences directly. Small and medium-sized manufacturers in Europe are also reporting that they are finding it difficult to find skilled workers with the necessary technical expertise—around half of them are already affected.

Manufacturing

These Two Megatrends Will Redefine the Industry by 2030

Another advantage of humanoid robots is that they can significantly reduce the strain on employees—for example when lifting heavy loads, performing monotonous, repetitive tasks or handling hazardous substances.

Why Humanoid Robots Find It So Difficult to Learn—And How AI-Generated Worlds Make Them Fit

For humans, moving and working safely in three dimensions is a matter of course—the result of millions of years of evolution. Machines, on the other hand, first have to learn these "rules of interaction". This is precisely one of the greatest challenges. At the center of this challenge is the "robot brain". Just like humans, robots also learn through experience: they have to understand how objects of different shapes and masses behave in different environments, how they keep their balance and avoid obstacles.

Such training data has hardly been available to date and has to be painstakingly created. Even for simple actions such as gripping a cup or climbing stairs, there are no ready-made data sets. Collecting such data is time-consuming and costly.

Transformation

Mobile Robotics as the Key to the Smart Factory

This is where synthetic data comes into play: with the help of physically precise digital twins, developers can train and test their AI models in virtual environments before they are used in the real world.

Another sticking point: the gigantic amounts of data must be available in a format that can be processed by robotics simulation programs. This is where the Universal Scene Description (OpenUSD) standard has become established. It forms the basis for creating, editing, visualizing and simulating virtual 3D worlds. On this basis, teams can generate realistic data, enrich it with photorealistic details and sufficient variety and thus develop models that can be flexibly transferred to a wide variety of real scenarios.

Without such synthetic data, it is impossible to train humanoid robots to behave like humans, such as walking, grasping or navigating in complex environments.

The Three-Computer Solution

Three closely linked computer systems are required to develop, test and execute the software of humanoid robots in real time—a major challenge that has now been overcome thanks to AI.

• Firstly: AI models analyze the generated data sets and extract the rules for the robots' movements and actions from them.
• Secondly: Robots can practise and refine their skills in simulated worlds.
• Thirdly: Modern robots have powerful on-board computers that make it possible to apply these rules and data in real time in the physical world.

This "three-computer solution"—consisting of robot brain training, data generation and simulation as well as real-time application—is what makes humanoid robots suitable for practical use. Developers around the world are already successfully using this approach to bring humanoid robots of various shapes and sizes into real-world use.

Ready for Use

Thanks to advances in AI, simulation and computing power, humanoid robots are finally becoming a reality. They are ready to operate in human environments, counteract the shortage of skilled workers and permanently change entire industries.

(kip)

Agricultural Engineering

Smart and Efficient: An Opportunity for Automotive Suppliers?