世界ロボット会議で「未来の高齢者介護の希望」を担う家庭用ロボットを囲みました

北京で開催中の世界ロボット会議では、人型ロボットの展示が絶対的な注目となっているスターダストインテリジェントのブースでは、AIロボットアシスタントS1がダルシマー、武道、書道の3大パフォーマンスを披露した。ある展示エリアでは、多数の専門的な聴衆とメディアが集まりました。

弾性ストリングでのエレガントな演奏により、S1 は、スピード、強さ、正確さによる繊細な操作と絶対的なコントロールを実証します。

CCTVニュースは、「書道」の背後にある模倣学習とインテリジェント制御に関する特別レポートを実施し、同社の創設者ライ・ジエ氏は、滑らかな動きの背後には、ハードウェア上での最高の力制御と最も人間に近い身体指標の追求があると説明した。 AI側では人間の実際の動きのデータが収集され、ロボットは強さに遭遇するとより強くなり、急速に進化することを学習します。

機敏でエレガントな詠春拳のボクシングテクニックは、観客から「ハードかつソフト」と評され、ロボットにサイバネティックなイップ・マンのスタイルを与えています。

ほんの3日前、この「AIロボットアシスタント」は、自宅で床を掃除し、猫に餌をあげ、ケーキを広げ、お茶を淹れており、ネチズンからは「サイバー執事」「将来の引退希望」と呼ばれていました。

とにかく、海外では人気があり、有名なFigure AI企業のCEOも目にしました。

ビデオで最も難しいカンフー茶を選択して、S1 のパフォーマンスを詳しく見てみましょう:

洗う、取る、淹れる、提供する、茶わんの使用まで、十数のプロセスから、コップ、ポット、カップ、スプーンなど、形、大きさ、材質、重さなど全く異なる物を掴んだり置いたりするための道具です。 go では、複雑な状況をリアルタイムに認識し、複雑なプロセスのインテリジェントな計画と実行を実証します。

力の制御は正確以上です - 磁器のカップからお茶を飲むために木のスプーンを取るとき、左手と右手で使用される力が異なり、陶器は滑らかで壊れやすいため、握ることができませんしかし、次の瞬間に金属製のやかんを持ち上げるときは、力の制御をリアルタイムで調整する必要があります。そうしないと、やかんを運んで水を注ぐことができなくなります。

細部はさらに擬人化されており、お茶をお渡しするときに、ティータオルを軽く叩いてわずかな水汚れを拭き取ります。これはお茶のエチケットに沿っています。

この驚くほど複雑な計画と正確な操作能力はネチズンから熱いコメントを集めました - Detroit: become human へ早送りします。

上記のスキルはすべて学習済みで、速度は 1 倍、リモコンなし (住み込みの乳母の後ろにリモコンはない)、実際の写真です。

その機敏さ、柔軟性、滑らかさは言うまでもなく、この優雅さは目に見えない形で動きに感情的な価値を加えます。

茶芸や楽器などの専門技術は人間が習得するのは容易ではなく、ましてや脳と体が二つに分かれたロボットなど、技術的な敷居はかなり高いです。

2010年に遡り、Appleの共同創設者スティーブ・ウォズニアックはロボットの知能を評価するためのコーヒーテストを提案しました。ロボットは、見知らぬ家に入り、キッチンを見つけ、道具 (コーヒーマシン、やかんなど) と材料 (コーヒー豆、砂糖、ミルクなど) を識別し、最後にコーヒーを淹れる必要があります。プロセス全体には 20 分もかかりません。

これは、未知の環境で移動し、物体を識別し、ツールや材料を操作し、人間のコマンドに従うロボットの能力をテストします。

誰かが世論調査を始めました、そのようなロボットは2040年に誕生しますか? 89% が賛成票を投じました。

カンフーティーの難易度やS1の進化速度を考えると、家庭用ロボットは身体化された知性に近いのでしょうか？

身体化された知性重視：価値の高い上半身

今年4月、S1は衣類のアイロンがけとたたみ、仕分けと仕分け、炒め物、掃除機をかけるなどの一連の作業に使用されました。清掃、競技用カップスタッキングなど。 めくるめく複雑な動作は、同仕様のロボットの中で「最強の動作性能」を第一印象とさせます。

8月にリリースされたS1は、人型の腰を備えているだけでなく、 実用的な車輪付きシャーシも備えており、住宅、オフィス、工場、ショッピングモールなど、人がいるほとんどの平坦なシーンをカバーできます。

二足歩行ロボットは物を拾うためにかがむことはできませんが、S1 は猫に餌をやるために極端にかがんでしゃがむことができます。

Whether it is scientific research or industry, upper limb operation is gradually becoming the focus of embodied intelligence. Judging from the history of the development of nature, those who can work with their hands are basically mammals with a very high level of intelligence. Most human operations are also completed by the upper limbs and hands. Intelligent decision-making + operation of the upper body is the core technical barrier to solving real needs and actual implementation.

The future of embodied intelligence: software and hardware integrated

Whether it is the "strongest operation", "highly human-like" or "silky-smooth movements", Stardust's What are the core technical barriers?

Founder Lai Jie mentioned in the interview that The secret is the system architecture integrating software and hardware, Design for AI.

On the one hand, it is the integration of hardware and software. The robot not only controls the accuracy of hard parts, but also grasps the strength of soft parts. Industrial robotic arms have high stiffness and can achieve high precision, but they cannot enter homes because they are not flexible enough. It is difficult for a human finger to move precisely one millimeter, but it can flexibly control the force, which is the key to interaction. From the structural design to the underlying hardware design, S1 will pursue the balance of hardness and softness.

The second layer is software and hardware,

that is, the strong coupling of AI and robots . AI relies on data and hopes to gain concepts and insights from it; while robots come from the pure physical world and understand multi-dimensional information such as position and force. Stardust allows the robot to be designed for AI, giving AI richer and higher-dimensional data (such as force and touch), so that AI can learn and execute more effectively.

For example, when peeling a cucumber, the S1 uses sensors to monitor changes in force during the transmission process. Like a human, it does not need to estimate the trajectory, but directly "feels" the force of the peeling knife on the cucumber, and then accurately controls the force output. This

unique rigid-flexible coupling transmission structure not only greatly improves the accuracy of movements, but also provides AI with rich data on "force".

If the cucumber is replaced by a carrot or a larger radish, it will be no problem. S1 can use "force" like a human and can handle things by analogy like a human.

In the research and development, Stardust drew on the team structure of Tencent RoboticsX Robotics Laboratory. Half of the people emphasized the use of AI algorithms for perception and motion control, while the other half focused on the robot body and continued to promote the "integration of software and hardware".

Multi-dimensional data builds the most powerful brain

In addition to "integration of software and hardware", data will also be the next watershed for embodied intelligence.

The core elements of AI development are algorithms, computing power and data. Computing power cannot be controlled, algorithms continue to evolve, and the only thing that remains constant is data.

A piece of trivia is: In the GPT era, there is almost zero directly available data to train robots.

When I came to Baidu, I realized the importance of data magnitude and data dimension, but the data dimension of robots has not improved for many years. At Stardust, he combined long-term thinking about "power" with AI to create a breakthrough embodied intelligent data acquisition solution.

Specifically, S1 can low-costly and efficiently utilize existing real-world video data and human motion capture data, and collect touch, force, vision, hearing, etc. from a first-person perspective

Multi-dimensional high-quality

Data.

As long as it interacts with the real world, S1 can

continuously generate new "learning materials", continue to evolve

, and move towards general artificial intelligence.

Safety and economy of robots

For robots to become human assistants, Laijie believes that there are two keys, one is safety, and the other is economy.

The safety of robots is not about grasping accuracy, but about grasping weight, and the strength of interaction with people and objects. This brings the foundation of safety back to "power".

Through rigid-flexible coupling hardware design and innovative power planning algorithm,

S1 is extremely safe

and can do no harm to people, yourself, or the environment during exercise.

On the price issue that everyone is most concerned about, relying on self-research of key components, S1 has obvious cost advantages. Laijie mentioned: "We hope that it will be the price of a car, and with the development of the entire industry, it will become a price that everyone can accept." This opens up a vast space and imagination for scientific research, commercial use and home use scenariosforce.

A typical application scenario is in biological, medical or chemical laboratories, which require high intellectual labor and cannot avoid repetitive, boring and precise experiments. At this time, a qualified "expert robot assistant" can work for professors accurately and day and night, liberating the labor force, unleashing creativity, and providing greater social value.

When S1 can learn, think and work like humans, interact smoothly and intelligently with humans, use human tools and equipment, and help humans complete boring, difficult or dangerous tasks. All of this will be warm and reflect the brand concept of "High Tech, High Life".

Start by employee No. 1 of Goose Factory Robotics Laboratory

Stardust Intelligence was established in Shenzhen in December 2022; the name comes from the ancient Latin proverb "Ad astra per aspera", which means "travel time" "Traveling Hard to Reach the Stardust" represents the company's long-term plan and firm commitment to the popularization of AI robotics technology.

The team spent a year carefully polishing the technology and products, and completed the S1 prototype in January 2024. It was first displayed in April and officially released in August, which continuously triggered industry discussions. Behind such efficient results is a team A strong team is working quietly.

Founder Lai Jie has more than 16 years of experience in robot research and development. He was the No. 1 member of Tencent RoboticsX Robotics Laboratory and led the development of the wheel-legged robot Ollie. Previously served as the person in charge of Baidu’s “Xiaodu Robot”.

Another founder, Dai Yuan, received a bachelor’s degree from UIUC and a PhD from UCLA, focusing on robot perception. She has published more than 30 papers in top journals such as Nature Comm and Science Adv, and holds more than 70 robotics patents. Like Laijie, she also joined the newly established RoboticsX in 2018.

The six core members of the start-up team are all from Tencent RoboticsX laboratory. The current team comes from major manufacturers and top universities such as Google, Huawei, DJI, and UBTECH.

The future of universal robots: filling gaps and serving families

Looking to the future, as the labor gap expands and the aging trend intensifies, universal robots have become a crucial solution. Some experts predict that by 2035, the world will need more than 100 million robots to fill vacancies in the workforce.

Not only that, the value of housework is often ignored, consuming a lot of our leisure time. Whether in industrial fields such as manufacturing and logistics, or in the home, the exploration of robot commercialization is driven by actual needs. However, in the long run, the introduction of robots into ordinary homes will be the most transformative application. Everyone has dreamed of having a "cyber nanny", just like Andrew in the movie "Robot Butler", who can do everything from cooking, cleaning to taking care of children.

Laijie said: The company’s vision is to empower billions of people with AI robot assistants. Whether it is taking care of the family or working in a factory, the more AI robots are like humans in learning, decision-making and execution, the more they can help people do more and better. It is also hoped that in the next five to ten years, AI robots will enter thousands of households.

The key question may be like the question that hopes to stimulate imagination at the end of every S1 presentation: "What else do you want S1 to do for you?"

以上が世界ロボット会議で「未来の高齢者介護の希望」を担う家庭用ロボットを囲みましたの詳細内容です。詳細については、PHP 中国語 Web サイトの他の関連記事を参照してください。