Robot Manipulation Assistance Using Learned Adaptive Control
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Users, especially those unfamiliar with robot manipulation or in unclear working environments, face challenges in accurately and efficiently controlling robots due to insufficient information about the working environment and potential delays between user operations and robot motions.
Innovation Solution
A robot control system that utilizes a learned model based on past robot manipulations to determine the degree of distribution associated with robot motions, sets a level of assistance, and generates an output command value to assist the user in achieving the desired robot state, combining input command values with predicted command values from both expert and user-specific models.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If a robot control system provides no assistance to the user, then the user maintains full control freedom, but the user struggles with accurate and efficient control due to insufficient information and delays
Solution Approach 1:
The patent introduces an assistance generation unit that acts as an intermediary between the user input and robot execution. This unit generates assistance command values based on learned models of past expert operations, mediating the control signal to provide guidance without completely overriding user intent. The assistance command is combined with the user's original command, creating a blended control signal that improves accuracy while maintaining user agency.
Solution Approach 2:
The system implements feedback by analyzing the difference between the current robot state and the target state, then using learned models to generate appropriate assistance. The assistance generation unit continuously monitors operation state differences and adjusts the assistance command values accordingly, providing adaptive feedback that guides the user toward more accurate and efficient operations based on historical expert data.
2Measurement precision
If the system provides strong assistance based on learned models, then manipulation accuracy improves, but the system may override user intent and reduce control flexibility
Solution Approach 1:
The patent applies partial action by generating assistance that is proportional to the operation state difference rather than fully overriding user commands. The assistance generation unit creates command values that partially correct user input based on the magnitude of deviation from desired behavior, providing just enough guidance to improve accuracy without completely taking control away from the user. This allows the system to intervene selectively rather than continuously.
Solution Approach 2:
The assistance level is made dynamic by continuously adjusting it based on the current operation state difference. The system adapts the amount of assistance provided in real-time, increasing assistance when the user's command deviates significantly from expert behavior and reducing assistance when the user's command is already close to optimal. This dynamic adjustment maintains control flexibility while improving precision when needed.
3Reliability
If the system uses multiple learned models (expert and user-specific), then the quality of assistance improves, but the computational complexity and processing time increase
Solution Approach 1:
The patent segments the learning model into two distinct components: an expert operation learning model and a user-specific operation learning model. Each model handles different aspects of the control assistance - the expert model provides ideal reference behavior while the user model adapts to individual user patterns. This segmentation allows the system to process and combine information from multiple sources without requiring a single monolithic complex model, making the overall system more manageable and interpretable.
Data Source
AI summary
A robot control system includes circuitry configured to: acquire an input command value indicating a manipulation of a robot by a subject user; acquire a current state of the robot and a target state associated with the manipulation of the robot; determine a state difference between the current state and the target state; acquire from a learned model, a degree of distribution associated with a motion of the robot, based on the state difference, wherein the learned model is generated based on a past robot manipulation; set a level of assistance to be given during the manipulation of the robot by the subject user, based on the degree of distribution acquired; and generate an output command value for operating the robot, based on the input command value and the level of assistance.