Multi-Skill Manipulation-Enhanced Mapping with Evidential Learning in Confined Environments

Yitian Shi, Nils Dengler, Jesper Mücke, Sicong Pan, Rania Rayyes, Maren Bennewitz

Keywords: Learning Representations for (inter)Active Perception, Uncertainty-Aware Control and Information Gain, Active System Identification and State Estimation, Active shape estimation

TL;DR: MS-MEM unifies active viewing, pushing, and grasping under evidential uncertainty modeling, allowing a robot to actively reduce occlusions and improve mapping accuracy while minimizing unnecessary disturbance to the scene.

Abstract: We propose Multi-Skill Manipulation-Enhanced Mapping (MS-MEM), a hierarchical evidential framework for uncertainty-aware occlusion mapping that jointly reasons over active viewpoint selection, non-prehensile pushing, and prehensile grasping. MS-MEM combines a scene-level metric-semantic belief map with a local grasp representation based on a full-evidential extension of vMF-Contact (FE-vMF), which models both grasp affordance and directional uncertainty. Within a POMDP formulation, the framework predicts the outcomes of candidate actions and evaluates them using Disturbance- and Occlusion-aware Information Gain (DOIG), a unified objective that balances expected visibility improvement against scene disturbance across heterogeneous action skills. In this way, MS-MEM enables localized and controllable occluder removal, improving visibility while minimizing unnecessary global disturbance to the scene. The demonstration video is at: https://www.youtube.com/watch?v=c32Qt-9ZIf44

Email Sharing: We authorize the sharing of all author emails with Program Chairs.

Data Release: We authorize the release of our submission and author names to the public in the event of acceptance.

Submission Number: 10