The system combines a compliant, hand-shaped gripper with built-in binary touch sensors. After initial contact, the robot uses tactile feedback to improve its alignment and assess whether its grasp is stable, rather than relying only on a pre-contact visual estimate.
In simulation, the method achieved more than 99% perching success across varied shapes and position errors of up to 0.6 meters and 50 degrees. Hardware tests covered 26 trials on diverse real structures and showed perching despite corrupted pose estimates.



