ManiSoft: Towards Vision-Language Manipulation for Soft Continuum Robotics

Researchers introduced ManiSoft, a benchmark and simulator for vision-language manipulation using soft continuum robot arms.

The suite includes four manipulation tasks plus automated scene and trajectory generation, with 6,300 generated scenes and expert trajectories.

Experiments show current models perform better in clean scenes than in randomized, cluttered ones.

ManiSoft is designed to test control of deformable robots in complex environments and help bridge the gap to soft-robot capabilities.