Researchers at Nvidia GEAR Lab and Carnegie Mellon University have developed a framework, ASAP (Aligning Simulation and Real Physics), that reduces the gap between simulated and real-world robot movements by 53%.

The system trains robots in simulation and then uses a specialized model to adjust for real-world variations.

In tests with the Unitree G1 humanoid robot, ASAP improved movement accuracy, enabling complex actions like jumps and kicks.

The system even replicated athletic movements of sports stars, the Decoder has reported.

While hardware limitations were encountered, ASAP offers a foundation for future advancements in natural robot movement, with the code available on GitHub.

Written by B.C. Begley