Field history

Physical AI timeline

Follow the progression from deep reinforcement learning and early world models to generalist robot policies and foundation models.

2015-2018

Foundations

Deep reinforcement learning, visual representation learning, simulation, and model-based control establish the technical base.

World Models

2019-2021

Early Embodied AI

Latent world models improve sample efficiency while large-scale embodied benchmarks and multimodal representations mature.

Dreamer: Reinforcement Learning with Latent Imagination DreamerV2 Ego4D: Unscripted First-Person Video Dataset

2021-2024

Egocentric and Cross-Embodiment Data

First-person video, synchronized ego-exo captures, and cross-robot datasets create the data layer needed for VLMs and VLAs.

Ego4D: Unscripted First-Person Video Dataset Ego-Exo4D: First- and Third-Person Skilled Activity Dataset Open X-Embodiment and RT-X DROID: Distributed Robot Interaction Dataset Mobile ALOHA: Low-Cost Whole-Body Teleoperation

2022-2024

Foundation Models for Robotics

Transformers, web-scale vision-language knowledge, cross-embodiment datasets, and open VLAs reshape general robot learning.

RT-1: Robotics Transformer for Real-World Control at Scale RT-2: Vision-Language-Action Models Transfer Web Knowledge to Robotic Control Open X-Embodiment and RT-X OpenVLA: An Open-Source Vision-Language-Action Model Genie: Generative Interactive Environments Voyager: An Open-Ended Embodied Agent with Large Language Models GAIA-1: A Generative World Model for Autonomous Driving pi0: A Vision-Language-Action Flow Model for General Robot Control

2025-Present

Physical AI Era

World models, humanoid foundation models, synthetic data, and generalist robot policies converge into integrated systems.

GR00T N1: An Open Foundation Model for Generalist Humanoid Robots NVIDIA Isaac GR00T N1.7 NVIDIA Isaac GR00T-Dreams Cosmos 3: Omnimodal World Models for Physical AI Genie 2: A Large-Scale Foundation World Model Genie 3: A General-Purpose World Model Gemini Robotics 1.5 V-JEPA 2: Self-Supervised Video Models for Physical Planning Figure Helix ASIMOV Benchmark for Robot Semantic Safety

2026

Production Safety for Physical AI

NVIDIA extends its autonomous-vehicle safety architecture to humanoids and industrial robots through Halos, Halos OS, IGX Thor, outside-in safety agents, and an inspection pathway for certification evidence.

NVIDIA Halos for Robotics