Robot Grip Detection Using Joint Torque and Shape Features
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing robot gripping technologies face challenges in accurately detecting object grip without tactile sensors due to high costs and size limitations, and methods like current measurement are unreliable due to friction-related errors.
Innovation Solution
A robot system that utilizes a state sensing unit to measure torque and shape features, producing a feature vector for a learning-based data sorter to determine if the hand is gripping an object, eliminating the need for tactile sensors by using a combination of shape and torque data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a tactile sensor is mounted to detect grip state, then measurement precision is improved, but device complexity and cost increase
Solution Approach 1:
The patent extracts the grip detection function from the tactile sensor and implements it through existing joint torque sensors and hand shape data. By taking out the detection function from a specialized tactile sensor and redistributing it across existing components, the system achieves accurate grip detection without adding complex sensor hardware.
Solution Approach 2:
The patent creates a virtual model of hand shape from joint angle data and uses this copied representation to infer grip state. Instead of directly measuring contact forces with tactile sensors, the system copies geometric information from joint positions and uses learning-based classification to determine grip status, eliminating the need for physical tactile sensors.
2Device complexity
If current measurement method is used to detect grip, then device complexity is reduced, but reliability deteriorates due to friction errors
Solution Approach 1:
The patent merges multiple data sources - joint torque values and hand shape information from joint angles - to detect grip state. By combining these complementary information sources through a learning-based classifier, the system achieves reliable grip detection that overcomes the limitations of single-method approaches like current measurement alone.
Solution Approach 2:
The system uses learning-based classification with trained models that incorporate feedback from training data to improve grip detection accuracy. The classifier learns from labeled training samples the relationship between joint torque, hand shape, and grip state, enabling reliable detection even in the presence of friction variations.
3Measurement precision
If tactile sensor is used for grip detection, then measurement precision is improved, but cost increases
Solution Approach 1:
The patent makes the existing joint torque sensors and hand model serve multiple functions - they are used not only for basic robot control but also for accurate grip state detection. This multi-functionality eliminates the need for separate tactile sensors, reducing system cost while maintaining measurement precision through intelligent data processing.
Solution Approach 2:
The system uses its own existing sensors and computational resources to perform grip detection without requiring external specialized components. The robot's joint torque sensors and processing unit serve the dual purpose of control and perception, making the system self-sufficient and cost-effective.
Data Source
AI summary
A robot, which may accurately detect whether or not the robot is gripping an object, even without a tactile sensor, as a result of using a sorter that utilizes a torque generated from a joint of a hand and the shape of the hand as feature data, and a control method for the robot includes a state sensing unit to sense a state of the robot, a feature vector producing unit to extract feature data from the state sensing unit and to produce a feature vector using the feature data, and a learning-based data sorter to judge an operating state of the robot using the feature vector produced by the feature vector producing unit and to output the judged result.