Never mind antiquated robot dogs. It was thrilling to witness backflipping humanoid robots performing kung-fu choreography at China Media Group's Spring Festival Gala show last month in Beijing. If the objective was to dazzle people at home and abroad with China's prowess in bleeding-edge robotic tech, Unitree, MagicLab, Galbot and Noetix did the job, despite tut-tuts from some Western reviewers saying that 'stage performance does not equate to industrial robustness'. We'd rather take Elon Musk's word when he had earlier said that 'people outside China underestimate China, but China is an ass-kicker next level' in humanoid tech.
There is reason why robots spray cars with paint a few microns thick, yet struggle to accomplish much simpler tasks such as plucking flowers. It's far easier to program a robot to do a specific task than infuse it with ordinary human capability. Plucking a flower would involve writing software for every muscle in the arm, wrist, hand and fingers to arrive at the required dexterity. The process becomes even more complicated if a robot has to choose one flower to pluck from a bush. The human brain processes information in real time to arrive at that decision. Robots don't yet. But all this is changing fast. General purpose robots are emerging from the shadow of their ancestors, and AI is bridging their learning gap. They can now analyse data in real time to refine behaviour. Better still, they are acquiring capability of learning how humans move and act. This opens up a large market for generalised robots where China has taken an impressive lead. As AI's foundational models become more sophisticated, robots could acquire motor skills to match, or exceed, humans.
Embodied AI can be far more disruptive in economic activity. A machine that can detect a faulty product and move it around to fix it is much more useful than a combo of a machine that detects and another that moves things around. Challenges remain in improving perception, behaviour, reasoning. But AI's making moves that would make Asimov sit up.
There is reason why robots spray cars with paint a few microns thick, yet struggle to accomplish much simpler tasks such as plucking flowers. It's far easier to program a robot to do a specific task than infuse it with ordinary human capability. Plucking a flower would involve writing software for every muscle in the arm, wrist, hand and fingers to arrive at the required dexterity. The process becomes even more complicated if a robot has to choose one flower to pluck from a bush. The human brain processes information in real time to arrive at that decision. Robots don't yet. But all this is changing fast. General purpose robots are emerging from the shadow of their ancestors, and AI is bridging their learning gap. They can now analyse data in real time to refine behaviour. Better still, they are acquiring capability of learning how humans move and act. This opens up a large market for generalised robots where China has taken an impressive lead. As AI's foundational models become more sophisticated, robots could acquire motor skills to match, or exceed, humans.
Embodied AI can be far more disruptive in economic activity. A machine that can detect a faulty product and move it around to fix it is much more useful than a combo of a machine that detects and another that moves things around. Challenges remain in improving perception, behaviour, reasoning. But AI's making moves that would make Asimov sit up.
