Many robot demonstrations look impressive. But behind the polished performances, some systems may still fall short of what can truly be considered AI.
Elon Musk has once again voiced skepticism about the current state of humanoid robots.
On October 10, U.S. local time, Musk joined the X Takeover technology gathering at Tesla’s Gigafactory Texas in Austin via a live video link. It was the first time the event had been held at a Tesla factory, with discussions centered around a broad question: “Where is the future of humanity heading?”
Yet Musk’s comments also highlighted a less glamorous reality: while humanoid robots are experiencing rapid growth, the technology is still far from mature.
01 | Focusing on Real-World Execution: Is Musk Pouring Cold Water on Humanoid Robots?
Rather than engaging in abstract discussions about the future of technology, Musk focused heavily on the practical challenges of robotics during the event.
When discussing the growing AI robotics race involving technology giants such as Google and NVIDIA, Musk questioned the significance of many seemingly flawless robot demonstrations.
In his view, a large proportion of robot demos rely on movements that have been pre-programmed or carefully choreographed in advance. Such systems may look highly capable in controlled environments, but that does not necessarily mean they possess genuine autonomous intelligence.
For Musk, the real test of an intelligent robot is not whether it can repeat a carefully prepared sequence of actions, but whether it can handle unexpected situations and operate autonomously in an unpredictable environment.
Musk also acknowledged that Tesla’s humanoid robot, Optimus, still faces significant technical limitations.
Tasks that appear trivial to humans—such as picking up a screw or fastening wooden components—can place demanding requirements on a robot’s force control, precision, hand-eye coordination and tactile sensing. A weakness in any one of these capabilities can cause the entire task to fail.
Musk also touched on Tesla’s plans for in-house chip development. He acknowledged that his team does not yet have mature semiconductor manufacturing capabilities, but said the goal is to develop some of the world’s leading chips.
The broader plan involves integrating chip design, packaging and testing into a more complete in-house process, with the ambition of designing and validating around 10 chip architectures or solutions each year.
Drawing on Tesla’s experience of building electric vehicles and SpaceX’s experience of developing rockets from the ground up, Musk argued that the same willingness to tackle difficult core technologies from scratch could also be applied to semiconductor development.
02 | Beyond the Hype: Is Practicality the Real Path Forward for Embodied AI?
Musk’s views on robotics and chips ultimately reflect a broader philosophy that has characterized his approach to technology entrepreneurship: focus on engineering, solve the hardest problems, and prioritize real-world execution over concepts and hype.
This may also be one of the most important challenges facing the global robotics industry today.
The AI robotics sector is increasingly at risk of becoming overly focused on demonstrations while paying insufficient attention to real-world deployment. Companies can optimize robots for standardized environments and carefully controlled scenarios, producing remarkably smooth demonstrations. But once these systems encounter the complexity and unpredictability of real-world environments, their performance can deteriorate significantly.
According to Stanford’s 2026 AI Index Report, humanoid robots can achieve a task success rate of 89.4% in standardized laboratory environments, while their success rate in performing basic daily tasks in real household environments falls to just 12.4%.
Meanwhile, more than 60% of humanoid robot shipments in China during the first half of 2026 reportedly went to research, testing and exhibition demonstrations, while fewer than 20% were deployed in stable production environments.
These figures highlight a fundamental gap between demonstration performance and practical intelligence.
The industry’s heavy focus on seemingly simple tasks such as screwdriving is therefore not accidental. These basic industrial operations expose some of the most important barriers to commercializing humanoid robots.
Industrial deployment requires robots to adapt to constantly changing environments, handle variations in objects and operating conditions, and maintain consistent performance without relying on carefully scripted movements.
To move beyond demonstrations, robots will need to overcome several fundamental challenges, including:
- Autonomous operation in unknown environments
- High-precision force control
- Advanced tactile sensing
- Robust perception and manipulation
- Generalization across different tasks and environments
Only by solving these problems can humanoid robots evolve from impressive demonstration platforms into genuinely useful productivity tools.
The rapid rise of humanoid robotics is undeniable. But if the industry wants to move from “robots that can perform a demo” to “robots that can reliably do a job,” there is still a long road ahead.
The next stage of the robotics race may therefore be less about making robots look impressive—and far more about making them reliably useful in the real world.