The Rise of Agentic Robotics: Beyond the 'Lego Era'
Robotics is currently in its 'Lego era,' but the emergence of 'Agentic Robotics' and foundation models for motion are bringing the industry closer to a transformative 'ChatGPT moment.'
Breaking the Data Bottleneck in Robotics
For years, robotics has been characterized as being in its "Lego era"—a fragmented landscape of actuators, sensors, and bespoke code that rarely translates from one machine to another. However, a new wave of "Agentic Robotics" is threatening to disrupt this status quo. By applying foundation models—similar to the ones powering LLMs—to physical motion, researchers are training generalized AI agents that understand how to move through space and time without being explicitly programmed for every task.
A primary challenge in robotics has always been data scaling. Unlike the internet, which offers a near-infinite supply of text for LLMs, high-quality robotic demonstration data is expensive and slow to acquire. Startups like General Intuition are now raising massive rounds—recently valued at $6 billion—to solve this by building models that can learn from a mix of human demonstrations, simulation, and even video data. This shift toward "agentic" systems means robots can plan, adapt to failures, and refine their own movements in real-time.
We are witnessing the transition from "automation" to "autonomy." While traditional robots follow a rigid script, agentic robots use their internal world models to solve problems. If a robot is tasked with picking up a box and its path is blocked, it no longer throws an error code; it assesses the environment, moves the obstacle, and completes the task. This adaptability is the key to moving robotics out of the factory floor and into the unpredictable world of homes and hospitals.
Source: IEEE Spectrum