Collision-Free Kinematic Posture Prediction Using 3D CNN Models
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for simulating collision-free postures for complex kinematic systems like robots and humans are time-consuming and require significant manual interaction or computational effort, limiting the effectiveness of human simulation technology.
Innovation Solution
A machine learning-based collision-free detection function (CFD) is trained using a dataset of 3D virtual environments, kinematic systems, and postures to predict collision-free postures, transforming the kinematic problem into an image classification problem solvable by convolutional neural networks (CNNs), thereby reducing computation time and manual interaction.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If methods such as Rapidly Exploring Random Trees (RRT) are employed to solve collision-free posturing, then collision-free postures can be found, but significant time is required to execute the solution
Solution Approach 1:
The patent pre-processes the 3D environment into a voxel grid representation before collision detection. This preliminary structuring of spatial data allows for much faster collision queries during posture evaluation, as the voxelized environment enables constant-time or near-constant-time collision checks compared to traditional geometric intersection algorithms.
Solution Approach 2:
The patent replaces traditional geometric collision detection algorithms with a voxel-based collision detection system. By substituting continuous geometric intersection tests with discrete voxel grid lookups, the system achieves significantly faster collision-free posture detection while maintaining accuracy.
2Measurement precision
If voxelization of the surrounding environment is performed for collision detection, then collision detection accuracy is improved, but time-consuming preparation is required
Solution Approach 1:
The voxelization of the environment is performed once as a preliminary step before simulation begins. The pre-processed voxel grid is stored and reused for all subsequent collision detection operations, eliminating the need to re-voxelize the environment for each posture evaluation or simulation step.
Solution Approach 2:
The patent uses a voxel resolution that is sufficient for collision detection accuracy but not excessively fine. By selecting an appropriate voxel granularity that balances accuracy with processing efficiency, the system achieves good collision detection performance without the computational overhead of ultra-fine voxelization.
3Reliability
If manual adjustment of individual joints in a kinematic chain is used, then collision-free postures can be achieved, but considerable manual interaction time cost is incurred
Solution Approach 1:
The system automatically evaluates collision-free postures by programmatically testing candidate postures against the voxelized environment model. This automated self-service approach eliminates the need for manual joint adjustment by operators, as the system independently determines which postures avoid collisions through algorithmic evaluation.
Solution Approach 2:
The patent replaces manual interactive adjustment with automated computational evaluation. By substituting human operator manipulation with algorithmic posture generation and collision checking, the system achieves collision-free posturing without manual interaction time costs.
Data Source
Figure 1
Figure 2
Figure 3~4
AI summary
System and a method for predicting a collision free posture of a kinematic system, the method comprising the following steps: a) receiving (201) a 3D virtual environment; b) receiving (202) a 3D representation of the kinematic system - called hereafter the 3D virtual kinematic system - and a set of 3D postures defined for the 3D virtual kinematic system; c) receiving (203) a target task to be performed by the kinematic system with respect to said surrounding environment; d) receiving (204) a prescribed location within said 3D virtual environment, said prescribed location defining a position at which the 3D virtual kinematic system has to be placed within said 3D virtual environment; e) applying (205) a collision free detection function - hereafter CFD function - trained by a machine learning algorithm to a set of input data comprising the 3D virtual environment, the target task, the prescribed location and the set of postures, wherein the CFD function is configured for generating as output a set of collision free postures enabling the kinematic system to perform said target task when located at said prescribed location; f) selecting (206) at least one collision free posture from said set; g) displaying (207) the 3D virtual kinematic system characterized by the selected collision free posture within the 3D virtual environment.