Act to Sense to
Act Better

About

Manipulation in unstructured environments demands more than accurate sensing—it requires purposeful interaction to acquire the right information at the right time. Active perception integrates sensing actions into manipulation: reorienting objects to expose features, sweeping a wrist to disambiguate geometry, or applying controlled forces to infer material properties. These strategies are essential for dexterity, enabling robots to adapt grasps, refine motion plans, and cope with uncertainty during execution.

This workshop focuses on three themes:

• Multi-Modal Active Perception: Purposeful tactile, visual, and multi-modal exploration to improve grasping, in-hand manipulation, and environment understanding.
• Algorithmic and Formal Contributions: Uncertainty-aware control, information-theoretic action selection, and learning-based methods that integrate sensing with acting.
• Integrated Systems and Applications: Combining active perception with regrasping, navigation, and active reconstruction for real-world tasks.