As with ChatGPT, it looks like the robotics sector is at a pivotal moment. The progress of robot intelligence is making the field more and more capable of dealing with everyday duties and operating in new environments with fewer human instructions, Unitree founder and CEO Wang Xingxing said Aug. 20. His remarks coincided with increased interest and excitement in humanoid robots in China, where numerous companies are pouring billions of dollars into creating robots that can interact with the real world.
Wang’s comments came after Unitree’s long-awaited listing in Shanghai attracted a huge number of investors. The company’s shares jumped significantly during the first day of trading and then were down 11% on the next day. The sudden swing of the current robotics market was apparent in both aspects of the robotics market, the excitement of what it means for the future, and the uncertainty as to how much time the prototypes are going to take to become a viable commercial product.
At a large robot conference in Beijing, the founder and CEO of the Hangzhou-based startup, Wang Xingxing, said that the company was heading towards a “ChatGPT moment” in the field of embodied intelligence.
The addition of OpenAI’s chatbot in late 2022 had a profound impact on the perception of AI, making the comparison with ChatGPT meaningful. Suddenly, large language models showed that machines can comprehend natural language and provide surprisingly complex answers. That breakthrough spurred more investment and research and adoption in industries. In robotics, there has never been such a dramatic event where a significant leap in software makes robots available to the masses.
What researchers are more and more calling embodied intelligence may be the next big advance. Unlike traditional AI systems that are mainly computer-based and digital, embodied AI enables machines to sense their environment, to take decisions, and to move and interact with their surroundings. For a humanoid robot, this translates to object recognition, instruction comprehension, room navigation, an ability to handle unforeseen scenarios, and to perform physical tasks without the need of a different program for each task.
Wang thinks the industry is getting closer to achieving that capability. He envisions a world where a robot would be able to go into a new house and obey a basic voice or written command without needing to be programmed or specially trained.
In the future, he said, the goal is to get a robot in an unknown home and get it to perform about 80% of the tasks via voice or text commands.
That goal represents one of the largest challenges on the robotics industry. Creating a machine with the ability to walk, balance, grab objects and do amazing things is just the beginning. The much more difficult challenge is to create software that can be used to enable the machine to understand the unpredictable real world and make decisions accordingly.
A robot performing a choreographed dance on a stage with controlled conditions. Things aren’t as simple in a crowded household. Furniture may move and objects may be placed in unintentionally strange locations, people act in ways that are hard to predict, and directions may be interpreted in multiple ways. Give “clean the kitchen”, and a machine must grasp what is meant by “clean”, determine where the objects are and how to clean them in a safe way.
This disconnect still lies at the heart of the industry. Since then, Unitree has emerged as one of the most prominent robotics firms in China, captivating the world with their robot dogs and humanoid robots showcasing synchronized dances and martial arts skills. The company is also making strides towards real-world applications, such as in industry. But the majority of its users are still at universities and research institutions, indicating the relatively fledgling nature of the technology.
Though the field is widely excited about the prospect of a breakthrough software, Wang has cautioned people not to take it for granted. He predicts that the level of intelligence of robots can significantly improve in two to three years in the best case. If the timeline is more conservative, then it takes 5 to 10 years, he proposed.
This cautious stance is critical as the financial market has certainly had more of a boom than the technology itself. Investors’ confidence in China’s humanoid robotics dreams were evident in the astounding success of its stock-market debut by Unitree. But just because a demonstration is successful or the share price is up, doesn’t make robots suitable for every home.
Industry watchers are also putting in their own estimates for when a breakthrough might occur. Chinese robotics firm Galbot founder Wang He believes the “ChatGPT moment” in the industry may be around the corner by 2028. He’s equally ambitious, though, in defining what robots might do: They’d have to be able to do about 70%–80% of ordinary tasks without needing to be specially trained for each situation.
By 2028, we can look forward to the ‘ChatGPT moment’ for embodied intelligence, he added, as more data is collected and new technologies come to fruition.
Data may have a significant role in getting there. Like the models used in AI, robots must learn from vast amounts of data.Robots, like AI models, require to learn from a large volume of information. Each successful movement, error, change in the environment, and human teaching or instruction can serve as a source of information for optimizing future movements. The challenge lies in gathering sufficient good real world data and learning in a safe manner for the robot.
