Robot Joint-Space Mapping Using Neural Networks for Reliable Positioning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for directing robots to specific locations are complex and prone to errors due to sensitivity to calibration and sensor measurements, and require cumbersome mathematical techniques and separate processes for perception and planning, making them unreliable and inflexible.
Innovation Solution
A system using machine learning to map joint space to Cartesian coordinates, employing a deep neural network trained via model-free reinforcement learning to generate transformations from task space to joint space, allowing for automatic handling of redundancy, joint limits, and obstacle avoidance without the need for differential equations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional mathematical techniques are used for robot positioning, then positioning capability is achieved, but system complexity increases and reliability decreases due to sensitivity to calibration errors and sensor measurements
Solution Approach 1:
The patent replaces traditional mechanical/mathematical positioning systems with a neural network-based system. The neural network learns the mapping from camera images to robot joint angles through training, eliminating the need for complex mathematical models, calibration procedures, and sensor fusion algorithms. This substitution reduces system complexity while improving reliability by making the positioning system robust to calibration errors and sensor noise.
Solution Approach 2:
The patent changes the fundamental parameters of the positioning system by transitioning from explicit mathematical models to learned representations. The neural network transforms input image features directly into joint angle commands through learned parameter mappings, rather than through complex kinematic equations and calibration parameters. This parameter transformation simplifies the system while maintaining positioning accuracy.
2Ease of operation
If complex mathematical techniques and separate perception-planning processes are used, then robot control is achieved, but ease of operation deteriorates
Solution Approach 1:
The patent merges the perception and planning processes into a single neural network. The network simultaneously processes camera images and outputs joint angle commands, eliminating the need for separate perception, localization, and planning modules. This unified approach simplifies operation while maintaining reliability by ensuring consistent coordination between perception and control actions.
Solution Approach 2:
The neural network serves multiple functions simultaneously: it performs image processing, feature extraction, coordinate transformation, and motion planning all in one model. This multi-functional approach makes the system easier to operate while maintaining positioning accuracy, as a single trained network handles the entire control pipeline without requiring separate calibration and configuration for each function.
3Adaptability or versatility
If hand-crafted kinematic and dynamic equations are used, then transformation accuracy is achieved, but adaptability decreases when robot configurations change
Solution Approach 1:
The patent performs preliminary training of the neural network on a diverse set of robot configurations and scenarios before deployment. During this training phase, the network learns to handle various robot configurations, link lengths, and operational conditions. This preliminary learning enables the system to adapt to configuration changes without requiring recalibration or reprogramming, while maintaining transformation accuracy through the learned representations.
Solution Approach 2:
The patent makes the positioning system dynamic by using a neural network that can adapt to different robot configurations through its learned parameters. Unlike static mathematical models that require recalibration when robot configurations change, the neural network dynamically adjusts its internal representations based on the training data, enabling it to handle configuration variations while maintaining accuracy.
Data Source
AI summary
Apparatuses, systems, and techniques to map coordinates in task space to a set of joint angles of an articulated robot. In at least one embodiment, a neural network is trained to map task-space coordinates to joint space coordinates of a robot by simulating a plurality of robots at various joint angles, and determining the position of their respective manipulators in task space.