In context
In-hand manipulation remains a key challenge in robotics, particularly re-orienting objects to arbitrary poses. While learning-based methods have advanced dexterous manipulation, many rely on high-degree-of-freedom hands or produce inefficient, non-human-like motions. This work addresses the gap by guiding five-fingered hands with different actuation levels to perform re-orientation tasks more naturally and efficiently.
What was reported
Researchers from the Beijing Institute of Control Engineering proposed a target position-guided policy generation method for in-hand object re-orientation. The method designs target positions based on actuation distribution, favoring the side with higher dexterity to enhance manipulation. The difference between actual and target object positions is incorporated into the reward function to keep the object near the target.
Inspired by human preparatory states, the method samples joint positions at each reset to improve manipulation capability. The policy is trained using proximal policy optimization (PPO) with a long short-term memory (LSTM) network and an asymmetric actor-critic architecture.
Simulations on three hands—9-DoA Schunk SVH, 13-DoA BICE, and 18-DoA Shadow—showed that the method enables each to approach predefined maximum consecutive success counts. Compared to methods without target position guidance, it significantly reduces the average number of steps required to complete re-orientation tasks.
Why it mattered
This work demonstrates that dexterous hands with varying actuation degrees can achieve efficient, human-like re-orientation through coordinated palm and finger actions. By improving learning efficiency and reducing steps, it advances practical dexterous manipulation for industrial automation, where precise in-hand handling is critical.
"The proposed method enables anthropomorphic five-fingered dexterous hands with different DoA to efficiently perform object re-orientation tasks in a human-like manner through coordinated action of the palm and fingers, significantly enhancing operational efficiency."
Source: 《机器人》期刊 (robot.sia.cn) · Published 2025-01-15 · “目标位置引导的五指灵巧手手内重定向”