The best carrier of embodied intelligence
Humanoid robots are widely believed by the industry to be expected to become the ultimate ideal form of robots, which are most similar to people in shape, can complete more and more flexible actions, and the interaction is more friendly. When the reporter interviewed the humanoid robot competition site, he heard the most sentence is, “Under the continuous empowerment of large models, humanoid robots will be the best carrier of embodied intelligence.”
PR6426/010-010+CON021/916-160 Embodied intelligence refers to the ability in the field of machine intelligence to interact with the environment and solve problems in a more natural and intelligent way by combining intelligent algorithms with the perception, actions, and environmental interactions of physical entities. As a general artificial intelligence embodiment carrier, humanoid robots can sense the physical world, control their bodies with multi-modal perception, and complete complex tasks when combined with large models of artificial intelligence.
“Data is very important for embodied intelligence. In this regard, humanoid robots have a natural advantage, and its data can be more directly obtained or transferred from human behavior data, which is a very big benefit.” Chen Jianyu said that in addition, there is a natural advantage that humanoid robots can better meet human emotional and aesthetic demands because they have a similar shape to humans.
“In the past, robot work may stay at the ‘cerebellar’ level, that is, how to make the robot walk more stable and the robot arm control more flexible.” It’s inherently a big disconnect from the business process.” Ji Chao, chief scientist of iflytek humanoid robot, explained in an interview with reporters that, for example, in the past, the robot walked from point A to point B, or grabbed some items on the assembly line, but it did not understand why to go to point B, and why to grab items, which is just the basic function of the robot. “If the robot only completes these basic functions, it can no longer meet the needs of business scenarios.”
PR6426/010-010+CON021/916-160 For example, if a power company has a gas leak or toxic gas leak, because of the transformer discharge, discharge will make the gas become highly toxic, so the staff need to enter the corresponding room, and then open the window, close the valve and other operations. This is a standard business process that allows staff to quickly understand what needs to be done when they encounter such a situation. However, such a scene, although the staff will wear chemical protective clothing and gas masks, but there are still potential risks to human safety and health.
“Therefore, we hope that humanoid robots can replace humans in potentially risky work scenarios, walk up to a valve with a gas leak, identify the valve, close the valve, and then open the window.” “The reason why this couldn’t be done in the past was because it [the humanoid robot] lacked a smart ‘brain’,” Ji said.
“We need to combine tasks with actual business processes, fine-tune them based on large models, and form large models based on multi-modal matrices under task scenarios. It allows the robot to make A steady series of actions based on our understanding of the scene, rather than simply walking from point A to point B, demonstrating its movement and grasping ability.” Ji Chao explained his expectations for the future of humanoid robots to reporters.