Dexterous Manipulation
contact-rich manipulation and in-hand manipulation
Overview
Dexterous manipulation in robotics involves precise control of objects, crucial for manufacturing, healthcare, and daily tasks. It includes both prehensile (grasping) and non-prehensile (non-grasping) methods, each with unique challenges. Non-prehensile manipulation, using simple end effectors like single rods, often faces under-actuation and complex planning constraints. To address this, we developed a method for object retrieval in cluttered environments using a rod-like pusher, combining an improved Rapidly-Exploring Random Tree (RRT) planner and a Model Predictive Control (MPC) scheme. This method’s contact-aware features enhance task feasibility and efficiency by enabling active obstacle removal and switching contact faces, validated through simulations and experiments.
Dexterous in-hand manipulation encounters difficulties with real-time contact discovery and inference. We propose a contact-implicit model predictive controller that generates real-time contact plans, allowing robust long-horizon in-hand tasks without pre-defined sequences. This approach enables efficient, large-displacement in-hand tasks and generalizes to various objects without pre-training. Additionally, we establish a comprehensive manipulation framework for deformable linear objects (DLOs) that includes large deformation modeling, whole-body shape control, and precise terminal manipulation. Our research addresses challenges like model uncertainty, high dimensionality, and multiple constraints, advancing robotic capabilities in complex, dynamic environments.
Winner of ICRA 2024 Robotic Grasping and Manipulation Competition - In-Hand Manipulation Track
The Robotic Grasping and Manipulation Competition - In-Hand Manipulation Track is focused on reconfiguring objects in robot dexterous hands. The competition tasks focus on two essential skills: in-hand precise manipulation and in-hand object re-orientation. Our solution involving an open-sourced Leap Hand, trajectory optimization, and reinforcement learning won the 1st place of the in-hand manipulation competition, and also won the Most Elegant Solution Award among all tracks in the RGMC.
Contact-Implicit Model Predictive Control for Dexterous In-hand Manipulation: A Long-Horizon and Robust Approach
(Jiang et al., 2024) [Website]
Inspired by recent advancements in contact-rich locomotion and manipulation, this paper proposes a novel model-based approach to control dexterous in-hand manipulation and overcome the current limitations.The proposed approach has the attractive feature, which allows the robot to robustly execute long-horizon in-hand manipulation without pre-defined contact sequences or separated planning procedures. Compared with other model-based methods, such a long-horizon feature enables replanning and robust execution of contact-rich motions to achieve large-displacement in-hand tasks more efficiently; Compared with existing learning-based methods, the proposed approach achieves the dexterity and also generalizes to different objects without any pre-training. Detailed simulations and ablation studies demonstrate the efficiency and effectiveness of our method. It runs at 20Hz on the 23-degree-of-freedom long-horizon in-hand object rotation task.
Contact-Aware Non-prehensile Robotic Manipulation for Object Retrieval in Cluttered Environments
(Jiang et al., 2023) [Website]
This paper proposes a new non-prehensile manipulation method for the task of object retrieval in cluttered environments, using a rod-like pusher. Different from existing methods, the proposed approach is with the contact-aware feature, which enables the synthesized effect of active removal of obstacles, avoidance behavior, and switching contact face for improved dexterity. Hence both the feasibility and efficiency of the task are greatly promoted. The performance of the proposed method is validated in a planar object retrieval task, where the target object, surrounded by many fixed or movable obstacles, is manipulated and isolated. Both simulation and experimental results are presented.
