Summary of Key Points
At the second World Humanoid Robot Competition, robots officially bid farewell to the “remote control era” and entered a phase of fully autonomous movement. Humans simply issue commands such as “run 100 meters,” and the robots use radar and cameras to navigate, plan their routes, and adjust their movements on their own. Although Tian Gong Ultra’s 100-meter time (8.64 seconds) surpassed that of Usain Bolt, it also revealed shortcomings, such as difficulties in turning and the need for cushioning to stop quickly at high speeds. The change in competition rules (all races to be fully autonomous) signifies a shift in the industry’s focus from “who can run the fastest” to “which system has the best perception, navigation, and control capabilities.” “Generalization ability” (the ability to adapt to different scenarios and tasks) is key for practical applications, but there is still a long way to go before these technologies are mature.
Detailed Explanation
1. Letting Go of Remote Controls: Robots Start to “Think” for Themselves
In the past, humans used remote controls to direct every step of a robot’s movement—lifting the left leg, stepping the right leg, adjusting the center of gravity, like controlling a puppet. Now, in fully autonomous mode, humans provide only a goal (e.g., “run 400 meters”), and the robot must accomplish three tasks:
- Finding Position: Equipped with radar and cameras, the robot can determine its location in real-time (similar to a smartphone’s navigation system).
- Planning the Route: It must follow the track and avoid obstacles.
- Adjusting Movements: Through simulation training, the robot develops the most suitable running posture (for example, Tian Gong Omni covers its face while running; this posture helps maintain balance and replaces the need for human arm swinging).
For instance, in obstacle races, robots like Zhi Yuan Ling Xi X2 use vision to detect steps and narrow passages and decide how to navigate without human intervention. This is like teaching a child to ride a bike—first with support, then independently, where the child must watch the road and maintain balance.
2. Turning and Stopping: Shortcomings in Autonomous Movement Are No Longer Hidden
Straight-line running is just the basics; turning and stopping are the real challenges:
- Difficulties in Turning: Moving in a straight line requires forward momentum, but turning involves handling lateral forces (similar to a motorcycle leaning when turning). Robots need to adjust their foot placement, body posture, and arm movements; any mistake can lead to a fall. With remote control, humans could correct these issues, but in autonomous mode, these shortcomings become apparent.
- Stopping Quickly: When slowing down from high speed, robots must estimate their speed and the friction of the surface and apply force in the opposite direction (similar to a person braking suddenly). In competitions, cushioning is used to allow robots to collide without slowing down, but in real-life applications, this is not feasible.
These issues indicate that robots’ “cerebellums” (the systems responsible for balance and gait) are not yet mature enough and need to develop similar flexibility to humans.
3. Competition Results on the Track ≠ Practical Use: The Gap Between Extreme Testing and Industry Applications
Competition on the track focuses on demonstrating maximum capabilities, while industry requires reliability and stability:
- Track Competitions: It’s okay for robots to collide with cushions or occasionally fall, as long as they run the fastest.
- Industrial Applications: Robots must operate reliably in factories, shopping malls, and homes. They cannot slip on tiles, get stuck on carpets, or crash into walls.
Competitions serve as a stress test to reveal potential problems (e.g., energy depletion or falls during turns), which need to be addressed in practical applications. For example, a robot that wins a 1500-meter race may not be able to work continuously for 8 hours in real life; improvements in cooling systems and battery life are necessary.
4. Evolving Competition: From “How Fast Can They Run” to “How Smart Are Their Systems”
Manufacturers used to compete by showing who had the fastest robots; now, the focus is on how well robots can solve problems independently:
- System Competition: Robots must not only run fast but also be able to turn, navigate obstacles, and stop autonomously, turning occasional successes into stable, repeatable capabilities.
- Generalization Ability Is Crucial: This refers to a robot’s adaptability—e.g., being able to sort different types of packages or assist with tasks without reprogramming. The industry is still in the first stage of generalization (e.g., sorting similar items) and will move on to the third stage, where robots can adjust their behavior based on the environment (e.g., slowing down in rain or automatically avoiding children).
Generalization ability reduces costs by eliminating the need to develop separate robots for each application, making them more widely useful in daily life.
Conclusion
Humanoid robots are transitioning from being novelty exhibits to practical tools. Fully autonomous movement is a significant milestone, but issues such as turning, stopping, and generalization ability still need to be resolved. Competitions like these are both showcases of technology and tests of practical feasibility. They highlight that for robots to truly replace humans, they must not only run fast but also be capable of thinking and adapting to various situations.