Robot Action Parameter Tuning Across Variable Object Postures
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for setting force control parameters in robots are inefficient, often resulting in overfitting, as they require repetitive trial-and-error processes under limited conditions, making it difficult to balance required force and task time in varying manufacturing and gripping scenarios.
Innovation Solution
A method that adjusts action parameters by determining evaluation positional postures through repeated task execution with varying object postures, comparing evaluation values to a reference, and iteratively updating action parameters based on task time and robot vibration until convergence.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If trial-and-error method is used to set force control parameter, then the parameter can be determined for specific conditions, but the parameter becomes overfitted and not suitable for actual operations with variations
Solution Approach 1:
The system performs preliminary actions by executing the task multiple times in advance with varied object positional postures. This preliminary execution phase allows the system to gather evaluation data across different conditions before actual operation, preventing overfitting by exposing the parameter optimization to diverse scenarios upfront.
Solution Approach 2:
The system dynamically adjusts the object positional posture during the evaluation phase, changing the starting position and orientation of the object across multiple trial executions. This dynamic variation ensures the force control parameter is optimized for a range of conditions rather than a single fixed state, improving adaptability.
2Reliability
If repetitive trial execution is performed to set action parameter, then suitable force and task time balance can be achieved, but the process is time-consuming and inefficient
Solution Approach 1:
The system implements feedback by measuring task execution time and robot vibration during each trial, then using this information to update and optimize the action parameter. The feedback loop continues until convergence is achieved, ensuring reliable parameter settings while reducing the number of trials needed through directed optimization rather than random search.
Solution Approach 2:
The system replaces manual trial-and-error adjustment with automated parameter optimization using computational algorithms. The control device automatically executes tasks, measures performance metrics, and updates parameters based on algorithmic optimization, substituting the mechanical trial-and-error process with an efficient computational system.
3Productivity
If force control parameter is optimized for specific conditions, then task performance is excellent under those conditions, but the parameter is not suitable for actual operations with manufacturing and gripping variations
Solution Approach 1:
The system applies local quality by optimizing the force control parameter for each specific object positional posture rather than using a single global parameter. By determining parameters tailored to local conditions (specific positions and orientations), the system achieves high performance for each configuration while the overall set of parameters handles variations robustly.
Solution Approach 2:
The system creates a universal parameter set that functions across multiple different object positional postures. By optimizing parameters for various positions and orientations during the evaluation phase, the resulting force control parameter becomes multi-functional, handling different manufacturing and gripping variations that occur in actual operations.
Data Source
AI summary
A method of adjusting an action parameter includes a positional posture determination step of making a robot execute a task a plurality of times in a plurality of positional postures different in positional posture of an object when starting the task to obtain evaluation values of the respective tasks, comparing the evaluation values of the tasks out of the evaluation values of the respective tasks with a reference evaluation value, and determining an evaluation positional posture from the positional postures in the tasks in which the evaluation value is no higher than the reference evaluation value, an updating step of making the robot operate with a tentative action parameter using the evaluation positional posture as a starting positional posture in the task to measure a time taken for the task or a vibration of the robot, and updating the tentative action parameter based on a measurement result, and a determination step of repeatedly performing the updating step until the time taken for the task or the vibration of the robot measured is converged to determine latest one of the tentative action parameters as an action parameter when actually performing the task.