콘텐츠로 건너뛰기
온라인 영업 엔지니어 · 09:00~21:00 CST 바오안, 선전, 중국
집/소식/로봇공학 및 AI
로봇공학 및 AI

Tencent는 파쿠르, 허들링, "한 인스턴스에서 다른 것들을 추론"할 수 있는 로봇 개 Max의 새로운 발전을 발표했습니다.

구스팩토리는 6월 14일 '로봇독 맥스'의 최신 진행 상황을 발표했다. "개 구멍"을 뚫을 수 있을 뿐만 아니라 장애물도 뚫을 수 있습니다. 현재 시중에 나와 있는 대부분의 로봇개들은 여전히 ​​가만히 서 있다.

Tencent는 파쿠르, 허들링,

구스팩토리는 6월 14일 '로봇독 맥스'의 최신 진행 상황을 발표했다. "개 구멍"을 뚫을 수 있을 뿐만 아니라 장애물도 뚫을 수 있습니다.

미니 로봇 개현재 시중에 나와 있는 대부분의 로봇개들은 여전히 ​​가만히 서 있다. Tencent의 새로 업그레이드된 로봇 개 Max는 일자리를 두고 실제 개와 경쟁하게 될 것입니다.

3c1b29f6-6fab-4048-9465-59e7c81d6dbe미니 로봇 개The new progress of this research was brought by Tencent Robotics X robotics laboratory. The experiment was established in 2018, and the robot projects currently launched include the robot dog Max 1st and 2nd generation, robot dog Jamoca, wheeled robot Ollie, self-balancing self-driving motorcycle, etc.

In 2021, Tencent officially released the first fully self-developed software and hardware multi-modal quadruped robot robot dog Max. In 2022, the second-generation quadruped robot "Max" was released. significantly improved. Tencent said that by applying the cutting-edge pre-trained AI model and reinforcement learning technology to the field of robot control, the flexibility and autonomous decision-making ability of the robot dog Max have been greatly improved.

미니 로봇 개2세대 4족 보행 로봇 '맥스'

By introducing the pre-training model and reinforcement learning technology, the robot dog can learn in stages, and effectively accumulate and store the skills and knowledge of different stages, so that the robot does not have to re-learn when solving new complex tasks, but repeats them. Use the learned attitude, environmental perception, and strategic planning knowledge at multiple levels to "infer other cases from one instance" and flexibly respond to complex environments.

Specifically, in the first pre-training study, Robotics X Lab cooperated with Tencent Games to use the motion capture data set in the game development process to let Max learn to walk, run, jump, stand and other actions.The second pre-training learning uses additional network parameters to connect the smart posture of the robot dog mastered in the first stage with the external perception, so that the robot dog can respond to the external environment through the smart posture that has been learned, and complete crawling and hurdle running.

, obstacle parkour shuttle and other difficult actions. In the third stage of learning, the additional network will obtain information related to complex tasks, such as in the game, obtain opponent information and flag information, and finally enable the robot dog to autonomously form cognition on the strategy side. To test these new skills that Max had mastered, the researchers designed a two-dog obstacle chase inspired by the obstacle chasing competition "World Chase Tag".

Based on the pre-trained model, the robot dog has not only learned the walking, running, jumping, standing and other actions of real dogs through deep reinforcement learning, but also has certain reasoning and decision-making capabilities. When the pursuer realizes that he can't catch up with the evader before he touches the flag, the pursuer will give up the pursuit, but will wander away from the evader, in order to wait for the next reset flag to appear.

In addition, when the chaser is about to catch the evader at the last moment, it will jump up and make a "pounce" action towards the evader. These are the active acceleration measures taken by the robot dog to ensure its own victory.According to reports, all the control strategies of the robot dog in the game are neural network strategies, which are learned in the simulation and passed through zero-shot transfer (zero-shot transfer), allowing the neural network to simulate human reasoning methods to identify never-before-seen new things, and deploy this knowledge to real robot dogs.

미니 로봇 개Tencent는 전체 프로세스가 가상 세계에서 교육되며 실제 기계 교육이나 조정이 필요하지 않다고 말했습니다. 로봇개 맥스는 앞으로도 계속 진화하고 적응해 나갈 예정이며, 수색구조 등 긴급작업에 투입될 것으로 예상된다.

At present, many companies have entered the field of quadruped robots. Boston Dynamics is famous for its robot dog "Spot". On August 10 last year, Xiaomi's first bionic four-legged robot CyberDog was officially unveiled. Xiaopeng Motors' ecological enterprise Pengxing Intelligent also released the world's first rideable robot on September 7 last year. Smart robot horse. In addition, there are a number of local start-ups such as Yunshen and Yushu Technology.

사전 훈련된 AI 모델과 심층 강화 학습 기술은 미래의 로봇이 다른 복잡한 작업을 해결할 수 있는 일련의 일반적인 솔루션을 제공하여 로봇이 실제 생활에 진입하고 인간에게 봉사할 수 있는 견고한 기반을 마련합니다. Tencent Robot Dog는 AI 기반 로봇을 생생하게 해석한 제품입니다.

AI가 만든 로봇과 기존 로봇의 가장 큰 차이점은 AI 로봇이 원래의 '자동화'에서 실제 '자율 학습'으로 옮겨가면서 인간과 유사한 지능적 특성을 어느 정도 보여준다는 점이다.

From a technical point of view, quadruped robots integrate the advanced level of bionic structure design of power system components, environmental perception technology, autonomous control technology, motion planning, motion control, SLAM and other technologies. With the improvement of robot performance and functions , machine vision can also be added to quadruped robots. It is foreseeable that with the continuous advancement of technologies such as computer vision, automatic control, artificial intelligence, gait planning and energy supply, the competitiveness of "robot dogs" will be further enhanced.

로봇개, 미니로봇개, 검사로봇개

귀하의 회선에 대해 이를 지정해야 합니까?부품, 목표 속도 및 국가를 보내면 엔지니어가 48시간 이내에 응답합니다.

견적 요청
관련된

더 많은 기사

TROY 어시스턴트영업 엔지니어 · 몇 초 만에 답변
안녕하세요. 저는 TROY의 영업엔지니어 봇입니다.
자동화에 필요한 것이 무엇인지 알려주십시오. 부품 중량, 도달 범위 또는 사이클 시간이면 충분합니다. 나는 귀하의 언어로 답변하고 48시간 안에 인간 엔지니어 앞에 귀하를 연결해 드릴 수 있습니다.
가격은 공개되지 않습니다. 48시간 이내에 구성별 견적 ·견적 양식