Humanoids and robot dogs navigate unseen terrains using attention mapping

Researchers developed a control system enabling humanoid and quadrupedal robots to navigate difficult terrain by using attention-based map encoding. The system helps robots focus on the most important areas for safe steps, allowing them to successfully handle unseen environments. Testing with ANYmal-D and Fourier GR-1 robots demonstrated impressive adaptability while requiring minimal power—just 25W on an onboard CPU—making it practical for various applications from disaster response to logistics.