Feng Jianfeng |Investment Advisors
a0380622060007|Practice number
This month, Tesla released the humanoid robot Optimus Gen-2**. It shows that Tesla's humanoid robot Optimus Gen-2 has the following changes compared with the previous generation of robot products:
1) Optimus Gen-2 has flexible hands, all fingers have a sense of touch, and can perform fine handling of objects, such as holding eggs between fingers;
2) A neck of 2 degrees of freedom allows the robot to turn its head within a certain amplitude;
3) Walking characteristics are similar to those of humans, and the walking speed suggests 30%;
4) At the same time, it can mobilize multiple joints of the whole body to maintain balance and complete an even 90-degree squat movement.
Since the beginning of this year, the research and development progress of Tesla's humanoid robot has accelerated significantly. The project was first unveiled in August 2021, and a year later the prototype was unveiled at AI Day. In May of this year, Optimus was able to walk and grasp objects smoothly, and in September, it was able to classify objects autonomously and was fully trained end-to-end by a neural network.
Tesla CEO Elon Musk even **, the demand for Optimus may reach 10 billion or even 20 billion units after it is mature and available. Although this goal seems very distant in the industry, the market generally believes that humanoid robots are expected to be practical in the next two years and begin to gradually enter the market.
One of the important catalytic factors is the empowerment of "embodied intelligence" of AI (artificial intelligence) technology.
Nvidia's AI system, Eureka, released in October this year, demonstrates the possibility of combining large language models (LLMs) with robot training.
It is reported that Eureka uses GPT-4 and generative AI to write software** to provide rewards for robots to perform intensive chemical Xi. Eureka can train robots to perform complex tasks, it teaches robots to quickly turn pens, as well as open drawers and cabinets, throw and catch balls, operate scissors and other tasks, Eureka can replace 80% of engineers' tasks, so that the average training efficiency of robots is increased by more than 50%.
Feng Jianfeng believes that AI determines the ceiling of humanoid robot product power, in the traditional deep Xi model, machine intelligence is often limited to specific scenarios, and under the empowerment of large models, robots can interact with the real world in multiple modalities, so as to achieve "embodied intelligence", and have the ability to evolve and apply across scenarios.
In recent years, "embodied intelligence" has gradually become a popular research direction, attracting research interests in many fields such as computer vision, natural language processing, and robotics. Since the first conference on robot learning (Corl) was held in 2017, we have witnessed the rapid development of the field of robotics Xi, including the emergence of a large number Xi of new intelligent tasks, algorithms, and environments. In the following years, especially at the CORL conferences in 2018 and 2019, a large number of embodied intelligence academic tasks began to be proposed and attracted attention, including embodied visual navigation, embodied question answering systems, etc.
In 2023, the CVPR 2023 Embodied Intelligence Symposium organized object rearrangement of AI Habitat, AI2-Thor, Igibson, and Sapien** devices, embodied Q&A, embodied navigation, and robot operation challenges. These embodied intelligence tasks have a completely different paradigm from other online AI tasks, that is, they use embodied agents (such as robots) to "see", "speak", "listen", "move", "reason" and other ways to interact with the environment and explore the task objectives, so as to solve various challenging tasks in the environment.
Feng Jianfeng believes that artificial intelligence will Xi and understand the cognitive paradigm of the human brain, and then approach human intelligence, which will be the general trendEmbodied intelligence, especially simulated human intelligence, is a feasible and efficient shortcut for artificial intelligence to move towards human intelligence.
Investment Advisory Profile:
Chief Investment Advisor of Vertex Finance, with more than 20 years of investment experience, independently developed the "3D Trading Operating System", and predicted the highs and lows of the historical Shanghai Composite Index for many times, which is loved by investors.
Disclaimer: This article does not constitute personal investment advice. The market is risky, investment needs to be cautious, please make independent judgment and decision-making.