In the realm of urban general-purpose intelligent robots, hardware differences are gradually diminishing. The true determinant of competitiveness lies in the capabilities of the AI “brain,” specifically in its ability to operate effectively in open environments. Recently, Liao Wenlong, the Chief Technology Officer of Coowa Technology, shared insights during an interview with various media outlets, including The Paper. Liao emphasized that the brain defines the upper limit of a system’s capabilities. To build an AI brain, it is essential to rely on AI for optimizing models and algorithms.
Founded in 2015, Coowa Technology is dedicated to achieving general artificial intelligence in the physical world through AI robotics. The company anticipates that its total product shipments will exceed 10,000 units in 2026, a figure that surpasses the total deliveries of the previous years, and it has already achieved a positive annual EBITDA (Earnings Before Interest, Taxes, Depreciation, and Amortization).
On February 5, Coowa officially launched the Coowa WAM 2.0 (World-Action Model), a universal world model base designed for high-frequency and standardized mobility and operational tasks. The Coowa WAM 2.0 is built on a Real-to-Sim-to-Real closed-loop model, enabling not just the ability to “see” the world but also to extrapolate within latent spaces, achieving deep coupling of drive and work.
Liao believes that the demand for physical AI models will experience explosive growth in the next 3 to 5 years. He envisions that in the coming five years, the physical front should strive for highly intelligent automation, while cloud-based decision-making will continue to require human-machine collaboration for some time.
Despite the current enthusiasm for the concept of physical AI, the industry faces significant challenges. Liao pointed out that one of the core issues is the infinite variety of long-tail scenarios in open environments and the scarcity of real data. “The key is not to exhaust all extreme cases but to equip machines with reliable and safe ‘zero-shot decision-making’ abilities under the premise of non-exhaustiveness,” he stated. This indicates that physical AI must follow a systematic path that maintains sustainable business viability, continuously gathering vast amounts of data from the real world and driving ongoing technological iteration through a scalable operational business loop.
In the context of smart city mobility services, Liao noted that “most of the massive data is ineffective and needs to be processed through automated pipelines for data mining, automatic labeling, incremental training, and simulation validation. The faster this infrastructure iterates, the stronger the competitive edge.” He argued that AI infrastructure determines data quality and scale, which in turn affects the upper limits of model capabilities. The core breakthrough for the next generation of embodied intelligence still lies in the continuous evolution of AI models.
Original article by NenPower, If reposted, please credit the source: https://nenpower.com/blog/explosive-demand-for-physical-ai-faces-data-scarcity-challenge/
