Robot Grasp Configuration Using Circular Anchors and Feature Maps
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing robot systems struggle to effectively utilize intermediate features from the technical field of the technical field of the technical field of the technical field of the technical problem of existing technologies have not effectively utilized the neural network, which are not utilized effectively utilized the intermediate features obtained by a neural network for more appropriate control of object manipulation with a smaller calculation amount.
Innovation Solution
The system employs a neural network with an encoder-decoder model to fuse intermediate features, uses circular anchors instead of box-shaped anchors, and learns the center and radius of circular anchors to reduce calculation load, enhancing learning efficiency and control precision.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional neural network approaches are used for object manipulation, then comprehensive feature extraction is achieved, but calculation amount increases and intermediate features are not effectively utilized
Solution Approach 1:
The patent segments the neural network processing into multiple stages: first obtaining intermediate features from early layers, then generating position heatmaps and grasp configuration heatmaps separately. This segmentation allows effective utilization of intermediate features without requiring the entire network to process all computations, thereby reducing overall calculation amount while maintaining grasping position accuracy.
Solution Approach 2:
The patent performs preliminary action by extracting intermediate features from the neural network before final grasp prediction. These intermediate features are then used to generate position heatmaps that guide the subsequent grasp configuration prediction. This preliminary extraction and utilization of features reduces the computational burden on later stages while improving prediction accuracy.
2Ease of operation
If traditional bounding box methods are used for region calculation, then object regions are defined, but rotation angle representation becomes complex and learning difficulty increases
Solution Approach 1:
The patent applies spheroidality by using circular anchors instead of traditional rectangular bounding boxes. The circular shape eliminates the need to represent rotation angles through complex bounding box orientations. Each circular anchor is defined by simple parameters (center coordinates and radius), making region calculation straightforward while naturally handling rotational symmetry without increasing learning difficulty.
Data Source
Figure 1
Figure 2
Figure 3~4
AI summary
According to one embodiment, an object manipulation apparatus (1, 2, 3) includes a feature calculation unit (3211), a region calculation unit (3213), and a grasp configuration (GC) calculation unit (3215). The feature calculation unit (3211) calculates a feature map indicating a feature a captured image of grasping target objects (104). The region calculation unit (3213) calculates, on the basis of the feature map, a position and a posture of a handling tool (14) by a first parameter on a circular anchor in the image. The handling tool is capable of grasping the grasping target object. The GC calculation unit (3215) calculates a GC of the handling tool by converting the position and the posture indicated by the first parameter into a second parameter indicating a position and a posture of the handling tool on the image.