“ Human robots continue to make technological breakthroughs, but can these breakthroughs get out of their own shell?
Over the past couple of years, human robots have been steadily warming up, with advances in technologies such as motors, smart hands, large models, and flexible skin. But behind the heated technological competition, a problem has gradually emerged: the robot essence, software hardware interfaces and development systems of different manufacturers are different, an algorithm changes the essence and often needs to be re-adapted; a scene is verified on one platform and not necessarily directly transferred to another platform.
It is not a question of the diversity of the robot itself, but of whether the different technological achievements can be effectively connected. Is it possible for robots to share the skills they learn from each other? Is it possible for one robot's accumulated experience to be the starting point for the improvement of another?
The physical environment of human societies, the means of production and the way in which they interact, are largely built around the human body and behavior. The long-term accumulation of human activities, tasks and collaborative experiences also provide a rich source of knowledge for machine learning.
In addition to continuing to breakthrough individual technologies, does the robotics industry need a common starting point based on a common human reference, capable of supporting technological development and application verification, and allowing decentralized technological achievements to form a common foundation for accumulation, migration and continuous evolution?
01.
The breakthroughs of the technological revolution, why do industries need a common benchmark?
The current competition in the robotics industry revolves largely around a single product: who controls the movement more steadily, who is more agile, and whose models are smarter.
These technological breakthroughs are important, but from the perspective of industrial development, it is equally worthwhile to consider whether they can be reused.
When robots, control interfaces, data formats, and development tools come together, developers not only have to solve application problems, but also have to pay a large amount of cost to adapt to different platforms. This limits the migration of technological outcomes and increases the burden of repeated R&D.
As a result, the industry needs not only better robots, but also a common foundation for reducing technology development, application migration and system integration costs.
So where should this foundation start?
02.
Why should people be the benchmark?
Human robots are of interest not only because of their shape, but also because of the large number of environments, tools, facilities, and workflows in human society that are built around human physical conditions and behaviors.
Doors, stairs, desk chairs, operating panels, and the way people interact with each other make up the existing social infrastructure. If robots want to access these environments directly and collaborate with human nature, it makes sense to refer to them as human beings.
Specifically, it can be understood from three dimensions.
Physical form is human-centered. Space, facilities and tools of human society, long-term development around human body structure, mobility and operating habits. The body size, joint layout and range of motion of robots should take fully into account the requirements of human activity space, existing facilities and tools, reduce the cost of transformation and adaptation of real-world scenarios as much as possible, and provide a physical basis for the learning and migration of human operating skills.
Interaction methods are based on human reference. They combine many modular interaction elements such as speech, expression, mouth type, vision, gesture and physical movement to achieve synergy of intent and behavior, making human-machine communication more natural, intuitive and efficient.
Ability to perform tasks in a safe, reliable and efficient manner, making full use of human knowledge, experience and data accumulated over a long period of time, learning continuously in real-world scenarios, iteratively optimizing, and gradually forming portable and replicable intelligence capabilities.
At the end of the day, it is not about unifying the appearance of robots on a human scale, but about building common references based on human physical form, interaction with practical experience, allowing robots to better use the environment and tools created by humans, learn the knowledge and skills that humans have accumulated, and lay the foundation for transnatural migration, experience accumulation and continuous evolution of intelligent capabilities.
03.
From baseline to industrial infrastructure
How do we translate human-based metals from design concepts to engineering practices?
One direction worth exploring is to build a reference robot platform that can support software hardware development, functional verification, and application exploration based on human physical conditions, interaction needs, and typical tasks.
The development of the smartphone industry provides an insight: the industry's prosperity comes not only from hardware advances, but also from the gradual evolution of the ecosystem of operating systems, development tools and application interfaces. The robotics industry cannot simply replicate mobile phone models, but it also needs to think about how to reduce development barriers and promote product replication.
In the field of humanoid robots, the reference body can be the starting point for this process.
On the one hand, through open interfaces and modular design, reducing the repetitive adaptation of application developers to underlying hardware details; on the other hand, through development tools, data specification and testing systems, making it easier for technological outcomes to be verified, compared and migrated. On the other hand, allowing robots the ability to form in different bodies, different tasks and different environments, to be effectively characterized, migrated and replicated, and driving robots from repetitive learning to experience accumulation, from single-machine intelligence to sustainable evolving intelligent systems. On this basis, in-house researchers, algorithm developers and industry users can collaborate around specific applications.
Of course, the benchmark kernel cannot automatically solve all standardization problems. There are still differences in the structure and controls of different robots, and migration across the kernel requires adaptation and verification.
Therefore, the benchmark body is not the entire infrastructure of the industry, but a physical carrier for building an open development system, accumulating reusable technological results.
04.
From concept to practice: exploration of real-world robot samples
As man-made robots gradually enter the home, public service and man-made collaboration scenes, physical interaction with nature will become an important part of the same intelligent system. The exploration of real-world robots is an attempt to combine physical interactions with familiar ways of interacting with humans to create a common physical platform that simultaneously supports operational intelligence and interactive intelligence.
If the idea is to be established, it will ultimately require testing of engineering practices.
Beijing Unilever Smart Technology Co., Ltd. has recently completed a real-world robot prototype that attempts to build a robotic platform based on human bodies, interactive needs, and operational capabilities to accommodate form, interaction, and task execution.
In its shape design, the model uses flexible skin wrapping throughout the body to explore the relationship between robotic appearance, touch and close human-to-human interaction.
In interactive design, the facial body has more than 30 degrees of freedom, allows for more expressive movements, and supports speech-language synchronization. This provides the foundation for further exploration of facial expressions, language communication and the combination of natural human-machine interaction.
In terms of operational capability, the sampler is equipped with an industrial-grade mechanical arm and 16 degrees of freedom. According to the project, its single-handed theoretical load of up to 5 kg provides the hardware base for tasks such as object operation and tool use.
The significance of this model should not be understood merely as the completion of a seemingly near-real-life robot. More importantly, it should be considered whether it can become a continuously evolving development platform: developing interactive systems, operating algorithms and application modules on it, validating it in real-world scenarios, and gradually forming replicable hardware components and development specifications.
From a sampling platform to an open industry platform, there is still a range of work to be done, including open interfaces, toolchain refinement, third-party development and scenario verification. But this offers at least a practical path to explore technology collaboration and application reuse from within.
05.
The benchmarks can go ahead, the standards need to be co-developed
Having a benchmark doesn't mean there should be only one robot in the industry.
Different heights, joints, drive modes and end executors may be suitable for different tasks. A truly valuable industrial ecosystem should allow for differentiated innovation while reducing collaboration between different systems as much as possible.
Therefore, the benchmark body is better suited for collaboration as an open starting point, rather than forcing all manufacturers to follow a single template.
An enterprise can lead the way in the exploration of benchmark platforms, open interfaces and development tools, and invite universities, research institutions, developers, and industry users to participate in verification. But real industry standards also need to be addressed in collaboration on issues such as interface compatibility, data sharing, test evaluation, and intellectual property rights.
These tasks cannot be done independently by one company, but need to be progressively advanced in engineering practice and multilateral collaboration.
The baseline can be explored ahead of time, development platforms can be opened up gradually, and compatibility standards require industry collaboration.
06.
Written at the end
Human robots still face many technological challenges such as energy supply, motion control, agility, environmental understanding, and autonomous decision-making. But technological breakthroughs are not all about industrial development.
As more and more businesses are able to manufacture humanoid robots, the industry needs to think about how to reduce the cost of iterative development, how to make technological achievements replicate across teams, how to make it easier for robots to get into the real world, and how to continue to accumulate capabilities in applications.
Using human benchmarks provides a worth exploring idea: using physical references to human physical conditions, using human interaction habits as a design basis, evaluating standards for real work tasks, and constructing benchmarks that can be developed, validated, and continuously generated.
It is not a unification of technological routes or an advance definition of industry standards, but is intended to provide a common starting point for decentralized R&D and application exploration.
Perhaps this is an important step that human robots need to take from the technological rush to industrial maturity.