Human Advancements in AI and Robotics Have Already Surpassed the Point of Creating Humanoid Beings
In the rapidly evolving world of technology, humanoid robots are making strides towards becoming a practical solution for physically demanding, repetitive, or dangerous tasks. However, transitioning these robots from impressive demonstrations to real-world functionality and commercialization is no easy feat.
One of the significant hurdles is the insufficient mobility and stability of humanoids. Walking remains unstable, and fine motor control, particularly upper limb manipulation, is limited. Despite early progress, such as robots walking a few steps in 2023 and running in 2024, the focus has recently shifted towards improving manipulation skills.
Another challenge is the lack of real-world training data. Robots largely rely on simulated or synthetic data, which often fails to capture real-world variability, leading to poor model performance. Collecting high-quality real-world data is expensive and complex, as demonstrated by companies like Tesla paying humans in motion capture suits to collect action data at $30 per hour.
Poor generalization and adaptability are also key issues. Current humanoids are often confined to narrow, specific tasks and struggle with flexibility or adapting to diverse, unforeseen scenarios. This lack of human-like versatility limits their practical application beyond controlled demonstrations.
Cost and manufacturing scale are also barriers to widespread deployment. Although manufacturing costs have dropped significantly, they remain a barrier. High production volumes and economies of scale are still emerging, with companies targeting tens of thousands of units in the near term.
Reliability for continuous operation is another challenge. For real-world deployment, robots must operate continuously without failure over extended periods, a feat that remains difficult due to technological immaturity. This reliability is critical for customers willing to pay a premium for dependable, round-the-clock service robots.
The humanoid robotics field is crowded and fragmented, with companies specializing in different aspects. No single player has fully solved safe, reliable, and affordable deployment challenges, making a unified path to commercialization complex.
Despite these challenges, humanoid robots offer the potential to ensure continuity in work processes without sacrificing safety, quality, or adaptability. In sectors with a combined payroll of $1.7 trillion in Europe, which are struggling to recruit younger workers, humanoids could prove to be a valuable solution.
AI must leave the confines of static datasets and controlled conditions to function in the real world, dealing with uneven floors, misplaced objects, unpredictable human behavior, and context-dependent nonverbal cues. The world is tailored to human scale, and humanoids must adapt to this reality to be functional.
At Humanoid, the goal is to build towards reliable autonomy through the use of teleoperation and end-to-end models. The aim is to build machines that can understand, navigate, and collaborate with the human world. Multimodal learning is essential for humanoids, as no single input channel is reliable enough to operate alone. AI systems need to be able to reason across modalities and timescales to integrate successfully.
The integration of humanoid robots into public life is a shift towards real-world applications. The real breakthrough in AI may come from integration in the physical world, learning to perform instead of merely predicting. The labor shortages in warehouses, assembly lines, and other industries are a global issue, particularly in Japan and Europe, where nearly 30% of the population is over 65.
Commercialization starts early at Humanoid, with testing in live environments and long-term partnerships around real use cases. Generalization is critical for humanoids, as they need to adapt their behavior when the environment changes. A humanoid robot named Shuang Shuang even received a diploma at a high school graduation ceremony in Fujian, China.
Modular architecture allows for independent iteration on subsystems while achieving system-wide coordination. Embodied reasoning, where language is grounded in space, time, and consequence, becomes more important than token prediction for humanoids. Human operators guide humanoids through key tasks in the early stages of development, and this hands-on data becomes the foundation for training new behaviors.
Regulatory framework is important, and teams should consider it from the start, especially in jurisdictions with stringent regulations like Europe. As humanoid robots become more integrated into our daily lives, it's crucial to ensure they are safe and beneficial for all.
With substantial investment, ongoing AI and hardware advances, and innovative training data strategies, the future of humanoid robots is promising. While fully flexible, robust humanoid robots for everyday commercial use are still emerging, the strides made so far suggest that they could be commercially available within the next two years.
Science and artificial intelligence can play crucial roles in overcoming the challenges faced by humanoid robots. For instance, AI can be employed to improve the robots' manipulation skills and adaptability, allowing them to handle diverse, unforeseen scenarios more effectively. Furthermore, space-and-astronomy technologies could be leveraged to enhance the mobility and stability of humanoids, enabling them to traverse uneven surfaces and maintain balance more efficiently. Technology, in turn, can aid in the development of cost-effective manufacturing processes and the creation of reliable, continuous-operation robots suited for commercialization.
