How robots learn: A brief, contemporary history

2026-04-17 · Source: MIT Technology Review · Field: Technology & Digital — Robotics & Autonomous Systems, Artificial Intelligence & Machine Learning · Depth: Intermediate, long

Summary

The robotics industry is experiencing a significant boom, with companies and investors injecting $6.1 billion into humanoid robots in 2025, quadrupling 2024 investments. This surge is driven by a paradigm shift in how robots learn, moving from rigid, rule-based programming to AI-driven methods. Early approaches, like OpenAI's Dactyl in 2018, utilized trial-and-error in simulations with "domain randomization" to adapt to real-world variations, though this method has limitations. The advent of large language models (LLMs) like ChatGPT in 2022 further catalyzed this shift, enabling robots to process vast amounts of data, including images and sensor readings, to predict actions. Google DeepMind's RT-2, released in 2023, trained on general internet images to enhance object interpretation, while Covariant's RFM-1, launched in 2024, allows for natural language interaction in warehouse settings. Companies like Agility Robotics are deploying humanoid robots, such as Digit, for tasks like moving shipping totes, demonstrating practical applications of these advanced learning techniques.

Key takeaway

For robotics engineers and product managers developing next-generation autonomous systems, you should prioritize integrating large AI models capable of processing diverse data types. Focus on iterative deployment to gather real-world data, which is crucial for refining robot performance and adaptability. This approach will enable more robust and versatile robots, moving beyond scripted actions to genuinely adaptive and interactive capabilities, despite the inherent risks of AI-generated actions.

Key insights

AI models and large datasets are revolutionizing robot learning, enabling adaptation and complex task execution.

Principles

• Simulated trial-and-error accelerates robot learning.
• Domain randomization improves real-world transferability.
• Large language models enhance robot perception and action.

Method

Robots learn by processing vast datasets (text, images, sensor readings) via AI models to predict next actions, often refined through simulated trial-and-error and real-world deployment for continuous learning.

In practice

• Use simulation with domain randomization for training.
• Integrate vision-language models for object interpretation.
• Deploy robots for data collection in target environments.

Topics

• Humanoid Robots
• Large Language Models
• Simulation Training
• Foundation Models
• Domain Randomization

Best for: Research Scientist, Investor, Entrepreneur, Robotics Engineer, AI Scientist, Director of AI/ML

Related on AIssential

• A robot operating system framework for using large language models in embodied AI — Connecting LLM agents to ROS enables versatile, robust, and scalable embodied AI through natural language.
• What is Physical AI? How Robots Learn & Adapt in Real Life — Physical AI integrates perception, reasoning, and action in the physical world, driven by advanced models and efficient compute.
• Why Robots Still Struggle With Simple Tasks (And What Might Finally Change That) | Karol Hausman, Co-Founder & CEO of Physical Intelligence — Combining internet-scale LLM knowledge with robot motion data enables general physical intelligence for complex tasks.
• From Seeing to Simulating: Generative High-Fidelity Simulation with Digital Cousins for Generalizable Robot Learning and Evaluation — Digital Cousins generates diverse, high-fidelity simulations from real-world panoramas for robust robot learning.
• This AI Watches Your Every Move — AI advancements are driving real-time coding, consumer wearables, and fully autonomous humanoid robotics, reshaping labor and production.
• Roundtables: Can AI Learn to Understand the World? — World models are emerging as a critical AI frontier to ground systems in real-world understanding, addressing LLM limitations in common sense and causality.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by MIT Technology Review.