Enhancing Dexterity in Robotic Manipulation via Hierarchical Contact Exploration

Xianyi Cheng,Sarvesh Patil,Zeynep Temel,Oliver Kroemer,Matthew T. Mason

2023-11-09

Abstract:Planning robot dexterity is challenging due to the non-smoothness introduced by contacts, intricate fine motions, and ever-changing scenarios. We present a hierarchical planning framework for dexterous robotic manipulation (HiDex). This framework explores in-hand and extrinsic dexterity by leveraging contacts. It generates rigid-body motions and complex contact sequences. Our framework is based on Monte-Carlo Tree Search and has three levels: 1) planning object motions and environment contact modes; 2) planning robot contacts; 3) path evaluation and control optimization. This framework offers two main advantages. First, it allows efficient global reasoning over high-dimensional complex space created by contacts. It solves a diverse set of manipulation tasks that require dexterity, both intrinsic (using the fingers) and extrinsic (also using the environment), mostly in seconds. Second, our framework allows the incorporation of expert knowledge and customizable setups in task mechanics and models. It requires minor modifications to accommodate different scenarios and robots. Hence, it provides a flexible and generalizable solution for various manipulation tasks. As examples, we analyze the results on 7 hand configurations and 15 scenarios. We demonstrate 8 tasks on two robot platforms.

Robotics

What problem does this paper attempt to address?

The paper attempts to address the problem of achieving high dexterity in robotic manipulation. Specifically, the authors propose a hierarchical planning framework (HiDex) aimed at tackling the following challenges:

1. **Non-smoothness due to contact**: When a robot interacts with the environment or objects, the contact introduces changes in system dynamics, making the planning problem non-smooth and complex.
2. **Diversity of fine motions**: Different manipulation tasks require different fine motions, making it difficult to simplify the problem through predefined primitive actions.
3. **Task-specific planners**: Existing manipulation planners are usually designed for specific tasks, making them hard to apply directly to other types of manipulation tasks.

To address these challenges, the HiDex framework uses a hierarchical Monte Carlo Tree Search (MCTS) method to decompose the planning of object trajectories, environmental contact modes, and robot contact points into three levels:

1. **First level**: Planning object trajectories and environmental contact modes. This level guides the object's motion by selecting contact modes, exploring external dexterity.
2. **Second level**: Planning robot contact points. Given the object trajectory, this level optimizes the sequence of contact points of the robot's fingers on the object's surface, exploring internal dexterity.
3. **Third level**: Path evaluation and control optimization. This level computes detailed path information, evaluates, and optimizes the path to generate the final control commands.

Through this hierarchical structure, the HiDex framework can efficiently handle planning problems in high-dimensional complex spaces, suitable for various dexterous manipulation tasks, including the intrinsic and extrinsic dexterity using fingers and the environment. The framework also offers flexibility and scalability, allowing it to adapt to different scenarios and robotic platforms with minimal adjustments.