Robot Command Generation From Taught Force and Pose Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing feedback control systems for robots performing assembly and pick-and-place tasks are not robust enough, especially when dealing with target objects of varying forms, due to inadequate definition and adjustment of sensor feature values to command values, and lack of force sensor employment.
Innovation Solution
A command value generating device that acquires and generates command values based on state data including action, position/orientation, and external force data during manual teaching, using an autoencoder structure to optimize internal parameters and select relevant state data for robust task execution.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If sensor feature values and command values are defined and populated in advance by a user, then the feedback control system can be configured, but it becomes difficult to determine whether a problem is one of definition, population, or adjustment
Solution Approach 1:
The patent implements a feedback mechanism where the robot executes teaching actions while sensors acquire state data, and this data is fed back to automatically generate or correct the mapping between sensor feature values and command values. This closed-loop feedback eliminates the need for manual definition and enables automatic problem detection through execution results.
Solution Approach 2:
The system performs self-service by automatically generating the mapping between sensor feature values and command values through executed teaching actions. The robot autonomously acquires state data during teaching and uses this data to establish the conversion relationships, eliminating manual configuration requirements and enabling self-diagnosis of problems.
2Ease of manufacture
If unmodified positions and force logging data from human assisted teaching are input as command values, then the system is simple to implement, but the force control system has low robustness
Solution Approach 1:
The patent transforms the raw position and force logging data into optimized command values by establishing conversion mappings based on actual teaching execution data. This parameter transformation process enhances the robustness of force control while maintaining implementation simplicity through automated data processing.
3Device complexity
If force sensors are not employed, then the system structure is simpler, but assembly and pick-and-place applications are not robustly executable
Solution Approach 1:
The patent uses state data from teaching actions as an intermediary to bridge the gap between simple sensor configurations and robust application execution. The acquired state data serves as a mediator that enables reliable assembly and pick-and-place operations without requiring complex force sensor systems.
4Reliability
If a generator is generated based on command values and state data from multiple teaching sessions, then task execution robustness is improved, but the system requires selection and optimization of state data segments
Solution Approach 1:
The patent segments the state data from multiple teaching sessions into relevant portions for generator generation. By dividing the comprehensive state data into manageable segments, the system can selectively process and optimize specific data portions, improving task execution robustness while controlling processing complexity through structured data organization.
Data Source
AI summary
An acquisition section (31) acquires command values to execute a task on a target object with a robot (40) and acquires state data representing a state of the robot (40) in a case in which an action of the robot (40) during the task is taught manually, which is state data of plural types including at least action data representing an action of the robot (40), position/orientation data representing a relative position and relative orientation between the robot (40) and the target object, and external force data representing external force received by the target object during the task. A generation section (33) generates a generator for, based on the command values and the state data acquired for corresponding times by the acquisition section (31), generating command values to execute an action with the robot (40) corresponding to the state data that has been input.