Robotic Grasp Planning via Finger Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Traditional methods for grasping complex three-dimensional objects using robotic hands are time-consuming and computationally intensive due to the need to independently control multiple degrees of freedom, making it difficult to achieve a secure and efficient grasp in real-time, especially with high-dimension robotic hands.
Innovation Solution
The approach decomposes the high-dimensional grasp planning into smaller, low-dimensional problems for each finger, using kinematic modeling to determine joint configurations and removing irrelevant object data, allowing for faster and more efficient grasp planning by focusing on feasible finger motions and contact points.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional iterative manual approach is used to plan a grasp, then the robot hand can achieve a secure grasp through manual adjustment, but the process is time-consuming and computationally intensive
Solution Approach 1:
The patent segments the high-dimensional grasp planning problem into multiple low-dimensional sub-problems, one for each finger. Each finger's planning is performed independently in its own low-dimensional space, avoiding the need to search the entire high-dimensional configuration space. This segmentation dramatically reduces computational complexity and planning time while still achieving secure grasps through coordinated finger positioning.
2Adaptability or versatility
If the robot hand has more degrees of freedom to achieve better grasp flexibility, then the adaptability of the grasp increases, but the difficulty of planning a feasible and robust grasp increases in reverse proportion
Solution Approach 1:
The patent divides the complex high-DOF hand into individual fingers, each with its own low-dimensional configuration space. By planning each finger independently in its reduced space and then combining the results, the system maintains the adaptability of high-DOF hands while reducing the planning complexity to manageable levels.
Solution Approach 2:
The patent transforms the high-dimensional grasp planning problem into multiple low-dimensional problems by changing the dimensionality of the search space. Instead of searching through all possible hand configurations simultaneously, the system searches through individual finger configurations in lower dimensions, making the planning process computationally feasible while preserving grasp flexibility.
3Reliability
If classic motion planning with collision detection is used to generate candidate grasps, then feasible grasps can be identified, but generating and evaluating thousands of candidate grasps is time and computation consuming
Solution Approach 1:
The patent segments the candidate generation process into finger-specific low-dimensional problems. Instead of generating thousands of full-hand candidate grasps through expensive collision detection, the system generates limited candidate configurations for each finger independently in its low-dimensional space, dramatically reducing the total number of candidates and computational requirements while maintaining grasp feasibility.
4Adaptability or versatility
If grasp planning is performed based on feedback from real-time vision systems and touch sensors, then the grasp can adapt to actual object properties, but the process is computationally expensive and time consuming
Solution Approach 1:
The patent performs preliminary grasp planning in low-dimensional spaces before execution. By pre-computing feasible finger configurations in reduced spaces and preparing the grasp plan in advance, the system reduces the computational burden during real-time execution, enabling faster response while maintaining adaptability through the structured planning approach.
Data Source
AI summary
A system, for planning a grasp for implementation by a grasping device having multiple grasping members, including a processor and a computer-readable storage device having stored thereon computer-executable instructions that, when executed by the processor, cause the processor to perform multiple operations. The operations include generating, for each of the multiple grasping members, multiple planar polygon representations of a three-dimensional object model. The operations also include transforming a planar polygon, of the multiple polygons generated, to a frame of a link of multiple links of a subject member of the multiple grasping members, forming a transformed polygon, being a cross-section of the object model taken along a member-curling plane of the subject member. The operations further include sweeping, in iterations associated respectively with each link of the subject member, the link from a fully-open position for the link to a point at which the link contacts the transformed planar polygon.