EVERY SIGNAL, SORTED.

South Korea, KAIST, Robotics, Quadruped Locomotion, Differentiable Simulation, Contact Dynamics, Policy Learning, Sim-to-Real Transfer, Hardware Evaluation and Safety Testing

KAIST taught a quadruped through contact in two minutes. Make the shortcut survive new ground

A KAIST team trained trotting and pacing policies from 100,000 simulated samples in 108 seconds, then moved them to a 45 kg robot. The impressive shortcut needs a terrain-by-terrain receipt before speed becomes trust.