Robot Force Estimation via Machine Learning Feedback
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing robot control systems fail to accurately control robot operations due to inadequate detection and classification of forces, leading to suboptimal performance in tasks like medical procedures where precise force feedback is crucial.
Innovation Solution
An information processing apparatus that employs a machine learning model to estimate external forces acting on a robot by combining force estimation values from sensors with external force response values, allowing for more accurate disturbance torque estimation and improved force control.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If force detection is performed using a force sensor and classification is performed using machine learning, then force classification capability is provided, but control accuracy is insufficient
Solution Approach 1:
The patent introduces a feedback mechanism where the estimated external force is fed back to the control unit to adjust the robot's operation. The control unit receives the estimated external force, compares it with the actual force sensor output, and adjusts control parameters accordingly. This closed-loop feedback enables the system to continuously improve control accuracy based on real-time force estimation feedback.
Solution Approach 2:
The patent introduces an intermediary estimation process between the force sensor and the control unit. Instead of directly using raw force sensor data for control decisions, the system uses a machine learning-based estimation unit as an intermediary that processes force information and provides refined estimates to the control unit. This intermediary layer enables more accurate control by filtering and enhancing the force data before it reaches the control logic.
2Device complexity
If only force sensor data is used for control, then system complexity is reduced, but control precision is insufficient
Solution Approach 1:
The patent merges multiple data sources and processing approaches into a unified control system. It combines force sensor data with machine learning-based estimation models, integrating sensor information with learned patterns from training data. This merging of sensor data and intelligent estimation algorithms enhances control precision while maintaining manageable system complexity through integrated processing.
Solution Approach 2:
The patent dynamically changes control parameters based on estimated external force conditions. The control unit adjusts control parameters such as gain factors, threshold values, and response characteristics in real-time according to the estimated force state. This dynamic parameter adaptation enables the system to maintain high control precision across varying operating conditions without requiring overly complex fixed control structures.
3Measurement precision
If machine learning model is used for force estimation, then external force estimation accuracy is improved, but computational requirements increase
Solution Approach 1:
The patent applies partial action by using machine learning models selectively rather than continuously. The estimation unit activates the machine learning model based on operational conditions, such as when force estimation is needed for control decisions. This partial activation approach reduces overall computational energy consumption while maintaining high estimation accuracy when required, avoiding the excessive computational burden of continuous full-model operation.
Data Source
AI summary
An information processing apparatus (100) includes an estimation unit (133) configured to estimate an external force estimation value which is an estimation value for an external force that acts on an object to be driven by performing a process based on a machine learning model with a force estimation value and an external force response value as an input, the force estimation value being an estimation value for a force that acts on the object to be driven, the external force response value being based on a value of an external force detected by a force sensor configured to detect the external force that acts on the object to be driven.