Robot Behavior Estimation Using Multi-Condition Skill Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The bilateral system in existing technologies lacks a mechanism to detect variations in conditions, making it difficult to perform object operations when environmental conditions change, as it assumes no variation between data storage and reproduction.
Innovation Solution
A behavior estimation apparatus that includes a collection unit to gather skill data under various conditions using a bilateral system with bidirectional control between a master and slave robot, and a behavior estimation device to estimate command values for the control target object based on this data, allowing the slave robot to adapt to changing conditions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If the bilateral system simply stores data and directly reproduces the data without detecting surrounding environment, then the system complexity is reduced, but the system cannot perform object operations when conditions vary
Solution Approach 1:
The patent introduces a surrounding environment detection mechanism that continuously monitors environmental conditions and feeds this information back to the data reproduction process. This feedback loop enables the system to detect variations in surrounding conditions and adjust data reproduction accordingly, resolving the contradiction between adaptability and system complexity by making the system responsive to environmental changes without requiring complete redesign
Solution Approach 2:
The patent transforms the static data reproduction process into a dynamic one by introducing environmental detection capabilities. The system now adapts its behavior based on real-time environmental conditions, allowing it to handle varying conditions effectively. This dynamic approach enables the system to maintain simplicity while gaining adaptability through conditional processing based on detected environment
2Extent of automation
If reinforcement learning is used to learn object operation skills, then the robot can determine behaviors through observation and reward mechanisms, but the number of attempts required becomes extremely large
Solution Approach 1:
The patent applies preliminary action by pre-collecting skill data under multiple different conditions before actual operation. Instead of learning through repeated attempts during execution, the system prepares comprehensive data sets in advance that cover various environmental scenarios. This preliminary data collection enables the robot to make accurate behavior determinations without requiring numerous real-time attempts, significantly reducing time loss
Solution Approach 2:
The patent uses copying by creating virtual representations of skill data from multiple conditions that can be referenced during operation. Rather than physically attempting operations repeatedly to learn, the system copies successful operation patterns from pre-collected data and applies them to current situations. This copying mechanism allows autonomous behavior determination without the time-consuming trial-and-error process
3Loss of time
If imitation learning is used to reduce the number of attempts, then the number of attempts can be significantly reduced, but the bidirectional property between operator and robot is not considered
Solution Approach 1:
The patent applies universality by collecting skill data under multiple different conditions and environments, creating a comprehensive data set that can serve various operational scenarios. This multi-functional approach ensures that the learned behaviors are not limited to specific conditions but can be reliably applied across diverse situations, thereby improving the success rate while maintaining the efficiency of imitation learning
Solution Approach 2:
The patent utilizes parameter changes by varying environmental conditions during data collection to capture a wide range of operational scenarios. By changing parameters such as environment type, object positions, and operational contexts, the system builds a robust model that accounts for bidirectional interactions between operator and robot. This approach improves reliability by ensuring the learned behaviors work across different parameter configurations
Data Source
AI summary
[Problem] Learning of object operation skills robust against variation of conditions is implemented.[Solution] A behavior estimation apparatus 100 includes a collection unit 200 configured to collect skill data obtained when a slave robot is operated under a plurality of different conditions by using a bilateral system capable of operating the slave robot via a master robot through bidirectional control between the master robot and the slave robot. The behavior estimation apparatus 100 further includes a behavior estimation device 300 configured to estimate a command value for causing the slave robot 520 to automatically behave, based on the skill data collected by the collection unit 200 and a response output from the slave robot 520.