Neuro-Capability Map Plug-Ins for Precise Robot Task Planning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing robot planning systems using voxelized capability maps face limitations in accurately representing complex reachable workspaces and performing precise actions due to discrete volumetric regions and abrupt transitions, which hinder efficient robot navigation and task execution.
Innovation Solution
A neuro-capability map plugin utilizing a neural network with semi-continuous or continuous resolution, trained to encode kinematic attributes and capable of high compression ratios, enabling efficient hardware-independent deployment and real-time inference with low energy consumption, allowing for precise robot actions and flexible monetization options.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If voxelized capability maps with discrete volumetric regions are used, then robot planning systems can evaluate reachable regions, but the representation accuracy and precision of complex workspaces deteriorate due to abrupt transitions
Solution Approach 1:
The patent replaces the traditional mechanical/volumetric voxel-based representation system with a neural network-based continuous representation system. The neuro-capability map plugin uses neural networks to continuously represent robot capabilities in workspace, eliminating the discrete voxel grid structure and its associated abrupt transitions, thereby improving representation accuracy while maintaining manageable complexity.
Solution Approach 2:
The patent changes the fundamental parameter of workspace representation from discrete voxel coordinates to continuous neural network output spaces. By transforming the representation from discrete volumetric regions to continuous capability probabilities and parameters, the system achieves smoother transitions and higher precision in representing complex reachable workspaces.
2Measurement precision
If high-resolution voxelized capability maps are used to accurately represent workspaces, then measurement precision improves, but storage requirements and computational resources increase
Solution Approach 1:
The patent substitutes the storage-intensive voxel grid system with a compact neural network model. Instead of storing capability information for every voxel in high-resolution grids, the system uses trained neural networks that can predict capability probabilities on-demand, dramatically reducing storage requirements while maintaining or improving resolution through continuous function evaluation.
Solution Approach 2:
The patent creates compressed representations of capability maps in the form of trained neural network models that can be stored and transferred. These model copies encapsulate the essential capability information without requiring the full-resolution voxel data, enabling efficient storage and deployment across different robotic systems.
3Productivity
If traditional voxel-based lookup tables are used, then robot planning can be performed, but task execution efficiency deteriorates due to discrete transitions and computational overhead
Solution Approach 1:
The patent replaces the discrete lookup table querying mechanism with continuous neural network inference. Instead of searching through voxel grids and handling discrete transitions, the system evaluates neural network functions that directly output capability probabilities, significantly reducing computation time and improving task execution efficiency through smooth continuous evaluations.
4Ease of operation
If discrete volumetric regions are used for capability representation, then robot navigation can be performed, but navigation smoothness and precision deteriorate due to abrupt transitions between voxels
Solution Approach 1:
The patent changes the representation from discrete voxel positions to continuous spatial coordinates evaluated through neural networks. This transformation eliminates abrupt transitions between discrete voxels and enables smooth, continuous navigation with higher position accuracy, as the neural network provides continuous capability probabilities across the entire workspace rather than at discrete voxel centers.
Data Source
AI summary
A component of a robotic system, including: processor circuitry; and a non-transitory computer-readable storage medium including instructions that, when executed by the processor circuitry, cause the processor circuitry to: train a neuro-capability map plugin, which is a continuous or semi-continuous resolution neural network component encoded with kinematic capability attributes with respect to an action to be performed by a robot in a workspace; and publish the neuro-capability map plugin to a robotic skills repository where it is obtainable by robotic controller circuitry for embedding within a neural network usable perform one or more inferences to control the robot to perform the action.