: Originally published in Science Robotics (2019), with related technical updates appearing in subsequent years (the timestamp 1641004453409 corresponds to early 2022, when this specific video clip gained viral traction on platforms like 4chan and Reddit).
: The paper describes a method using NVIDIA Isaac Gym and deep reinforcement learning to train legged robots in a simulation that is so fast and accurate that the robot can "learn" complex locomotion in minutes of real-world time. Why this file name is famous 1641004453409.webm
: Learning to Walk in Minutes Using Massively Parallel Deep Reinforcement Learning : Originally published in Science Robotics (2019), with
The specific filename 1641004453409.webm became a recurring "meme" or "reference file" on imageboards because it demonstrates the "zero-shot" transfer from simulation to reality—where the robot mimics the exact flailing and recovery movements it practiced in a virtual environment to get back on its feet. The video specifically showcases the robot (developed by
The video specifically showcases the robot (developed by ANYbotics and ETH Zurich) learning to walk and recover from falls in a simulated environment before being deployed to the real world. Paper Details
: Jemin Hwangbo, Joonho Lee, Alexey Dosovitskiy, Dario Bellicoso, Vassilios Tsounis, Vladlen Koltun, and Marco Hutter.