Panoramic sensing offers wide field-of-view coverage, yet 3D reconstruction from sparse panoramas remains challenging under rotation-dominant, weak-parallax motion. In such regimes, SfM/SLAM initialization is often ill-conditioned and unreliable. We present PanoImager, an SfM-fre…
Exploiting the morphological symmetry of robotic systems, such as humanoid and quadruped robots, is a promising direction for improving robot learning. In deep reinforcement learning (DRL) for robot control, prior studies have leveraged such symmetry to improve learning efficienc…