Utilizing Human-Inspired Dexterous Picking Skills for Enabling Context-Aware Robotic Manipulation in Multi-Object Scenarios
Abstract: This research aims to emulate the dexterity and precision found in human grasping capabilities, particularly when dealing with difficult-to-pick objects and challenging manipulation scenarios. To solve these challenges, we leverage four dexterous picking skills inspired by human manipulation techniques that include sliding, pushing to a vertical surface, leveraging a horizontal surface, and flipping objects. The proposed approach extends beyond traditional methods by incorporating a decision-making process that assesses which, where, and how to apply specific manipulation skills for objects within the scene. Utilizing deep neural networks, the system identifies the most suitable manipulation skill for each object in the scene, assigns confidence scores indicating the potential success of each pick, and predicts precise skill locations. The adaptability of the proposed system is rigorously evaluated through a series of real-world experiments, encompassing scenarios involving known, unknown, and occluded objects. These experiments, comprising 45 trials with 150+ grasps, validate the system’s reliability and robustness, particularly in cluttered settings. This research helps bridge the gap between human and robotic grasping, showcasing promising results in various practical scenarios.