Topic
4 articles
Three of the most powerful technology organisations in the world — NVIDIA, Google DeepMind, and Disney Research — jointly built and open-sourced Newton, a physics engine for training robots. It runs 70x faster than existing simulators. Here is why it matters.
Every robot that learns in simulation and deploys in the real world has to solve sim-to-real transfer. It is the reason robots fail when they leave the lab, and solving it is what NVIDIA, Google DeepMind, and Disney's Newton physics engine is actually attempting.
The cute rolling robots at Disney theme parks are not props and they are not remote-controlled. The BDX droids are trained using the same reinforcement learning and physics simulation techniques as industrial robots. Here is how they work and what makes them technically remarkable.
At the CCTV Spring Festival Gala — watched by nearly one billion people — Unitree's humanoid robots performed autonomous kung fu, aerial flips, and parkour. Last year they barely walked. Here is what actually happened and what the technical progress means.