The Robotics Renaissance: Navigating the Clash of Hardware and AI
December 6, 2024, 3:46 pm
2024 marks a pivotal year for robotics. The landscape is shifting, and a new generation of startups is emerging. But beneath the surface, a battle brews. It’s a clash of ideologies—hardware versus software, engineers versus AI experts. This pecking order shapes the future of robotics.
In the robotics arena, hardware teams often feel like the underdogs. They toil away, building the physical bodies of robots—humanoids, quadrupeds, and robotic arms. Yet, they find themselves overshadowed by software-first startups. These firms, often led by veterans of artificial intelligence, wield the power of large models and advanced algorithms. The divide is palpable. Hardware teams are viewed as the builders, while software teams are seen as the brains.
Wang Sheng, a partner at Innoangel Fund, categorizes these startups into three groups. The first is hardware-first teams, the backbone of robotics. The second group consists of AI veterans pivoting into robotics, leveraging their expertise in computer vision and reinforcement learning. The third, a smaller elite group, specializes in large models, often regarded as the pinnacle of the hierarchy.
This hierarchy isn’t just a matter of pride; it reflects deeper industry tensions. Many hardware companies in China shy away from AI, opting instead for open-source models. The cost of developing proprietary AI is daunting. Unitree Robotics exemplifies this trend. Their founder emphasizes that robots are their foundation, even encouraging customers to swap out software while keeping their hardware intact.
The debate over integrating intelligence into robotics reveals a fragmented industry. Hardware representatives express frustration over the lack of consensus on the “soft” aspects of robotics. What constitutes a robotic brain? How should embodied intelligence be constructed? These questions linger, leaving the industry in disarray.
As software-focused firms begin to develop their own hardware, the ecosystem becomes even more fractured. Investors joke that today’s hardware companies resemble video production teams, creating elaborate setups for staged demos. These demonstrations often create an illusion of artificial general intelligence. In reality, they mask the technical challenges lurking beneath the surface. A slight change in the environment can derail a robot’s performance.
The question arises: if large models thrive in smartphones and computers, why do they struggle in robotics? The answer lies in the limited application of AI among hardware companies. They typically rely on general-purpose language models, which lack the spatial intelligence crucial for robotics. The result? A reliance on massive datasets that introduce hallucinations, disrupting task execution.
No Chinese team has yet developed large models optimized for robotics. A workaround involves integrating a “cerebellum layer” between multimodal large models and robotic hardware. This layer breaks tasks into manageable subtasks. For instance, making coffee becomes a series of steps: “grab a cup,” “grind beans,” “pour water.” While this coordination allows robots to execute tasks, it introduces new challenges. Complex operations require extensive predefined subtasks, and data scarcity remains a significant hurdle.
Many entrepreneurs entered the robotics field with high hopes for large models, only to confront significant gaps. The fragmentation has become unsustainable, prompting a collective shift in industry thinking. By late 2024, investment trends began to change. Investors who once equated robotics with humanoid hardware are now pivoting toward embodied intelligence. Companies like Unitree and Zhiyuan Robotics, once valued over USD 1 billion, are now seen as too expensive for most investors.
Chinese startups face steep domestic fundraising challenges, forcing them to rethink their narratives. The market feels cold for pure hardware players. Recent funding rounds confirm this shift. Skild AI and Physical Intelligence have seen valuations soar past USD 10 billion globally. In China, Galaxea AI secured investment from Ant Group, while X Square and Qianjue Technology closed significant rounds.
Wang notes that investors are increasingly focused on embodied intelligence, recognizing its potential to drive generalized robotic tasks. Even hardware manufacturers, once skeptical of generalization, are now exploring foundational models to build specialized capabilities. The robotics industry is in a state of chaotic consensus. Yet, one truth stands out: the future lies in the integration of hardware and embodied intelligence. Neither can thrive in isolation.
The race to achieve higher business efficiency is on. Companies must master both AI and hardware. More importantly, they must foster mutual respect. The clash of ideologies is not just a battle for supremacy; it’s a quest for collaboration. As the industry evolves, the lines between hardware and software will blur. The ultimate goal is clear: create robots that can navigate the complexities of the real world.
In this robotics renaissance, the stakes are high. The integration of hardware and AI is not merely a trend; it’s the foundation for the future. As startups navigate this complex landscape, they must embrace the challenges and opportunities that lie ahead. The future of robotics is not just about building machines; it’s about creating intelligent systems that can learn, adapt, and thrive in an ever-changing world. The race has begun, and the finish line is a new era of robotics.
In the robotics arena, hardware teams often feel like the underdogs. They toil away, building the physical bodies of robots—humanoids, quadrupeds, and robotic arms. Yet, they find themselves overshadowed by software-first startups. These firms, often led by veterans of artificial intelligence, wield the power of large models and advanced algorithms. The divide is palpable. Hardware teams are viewed as the builders, while software teams are seen as the brains.
Wang Sheng, a partner at Innoangel Fund, categorizes these startups into three groups. The first is hardware-first teams, the backbone of robotics. The second group consists of AI veterans pivoting into robotics, leveraging their expertise in computer vision and reinforcement learning. The third, a smaller elite group, specializes in large models, often regarded as the pinnacle of the hierarchy.
This hierarchy isn’t just a matter of pride; it reflects deeper industry tensions. Many hardware companies in China shy away from AI, opting instead for open-source models. The cost of developing proprietary AI is daunting. Unitree Robotics exemplifies this trend. Their founder emphasizes that robots are their foundation, even encouraging customers to swap out software while keeping their hardware intact.
The debate over integrating intelligence into robotics reveals a fragmented industry. Hardware representatives express frustration over the lack of consensus on the “soft” aspects of robotics. What constitutes a robotic brain? How should embodied intelligence be constructed? These questions linger, leaving the industry in disarray.
As software-focused firms begin to develop their own hardware, the ecosystem becomes even more fractured. Investors joke that today’s hardware companies resemble video production teams, creating elaborate setups for staged demos. These demonstrations often create an illusion of artificial general intelligence. In reality, they mask the technical challenges lurking beneath the surface. A slight change in the environment can derail a robot’s performance.
The question arises: if large models thrive in smartphones and computers, why do they struggle in robotics? The answer lies in the limited application of AI among hardware companies. They typically rely on general-purpose language models, which lack the spatial intelligence crucial for robotics. The result? A reliance on massive datasets that introduce hallucinations, disrupting task execution.
No Chinese team has yet developed large models optimized for robotics. A workaround involves integrating a “cerebellum layer” between multimodal large models and robotic hardware. This layer breaks tasks into manageable subtasks. For instance, making coffee becomes a series of steps: “grab a cup,” “grind beans,” “pour water.” While this coordination allows robots to execute tasks, it introduces new challenges. Complex operations require extensive predefined subtasks, and data scarcity remains a significant hurdle.
Many entrepreneurs entered the robotics field with high hopes for large models, only to confront significant gaps. The fragmentation has become unsustainable, prompting a collective shift in industry thinking. By late 2024, investment trends began to change. Investors who once equated robotics with humanoid hardware are now pivoting toward embodied intelligence. Companies like Unitree and Zhiyuan Robotics, once valued over USD 1 billion, are now seen as too expensive for most investors.
Chinese startups face steep domestic fundraising challenges, forcing them to rethink their narratives. The market feels cold for pure hardware players. Recent funding rounds confirm this shift. Skild AI and Physical Intelligence have seen valuations soar past USD 10 billion globally. In China, Galaxea AI secured investment from Ant Group, while X Square and Qianjue Technology closed significant rounds.
Wang notes that investors are increasingly focused on embodied intelligence, recognizing its potential to drive generalized robotic tasks. Even hardware manufacturers, once skeptical of generalization, are now exploring foundational models to build specialized capabilities. The robotics industry is in a state of chaotic consensus. Yet, one truth stands out: the future lies in the integration of hardware and embodied intelligence. Neither can thrive in isolation.
The race to achieve higher business efficiency is on. Companies must master both AI and hardware. More importantly, they must foster mutual respect. The clash of ideologies is not just a battle for supremacy; it’s a quest for collaboration. As the industry evolves, the lines between hardware and software will blur. The ultimate goal is clear: create robots that can navigate the complexities of the real world.
In this robotics renaissance, the stakes are high. The integration of hardware and AI is not merely a trend; it’s the foundation for the future. As startups navigate this complex landscape, they must embrace the challenges and opportunities that lie ahead. The future of robotics is not just about building machines; it’s about creating intelligent systems that can learn, adapt, and thrive in an ever-changing world. The race has begun, and the finish line is a new era of robotics.