Robot Gripping Position Inference Using Multi-Position Regions

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing robot systems face precision issues in calculating gripping positions for workpieces due to multiple correct answers for input data, leading to degraded machine learning and inference precision, especially when imaging workpieces with cylindrical or differently sized portions.

Innovation Solution

An information processing apparatus and method that utilize a learned model to process sensing data from a vision sensor, generating images that indicate possible gripping positions for robot fingers, thereby improving inference precision by considering regions rather than individual positions, and calculating holding positions in three-dimensional space.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Adaptability or versatility

If multiple gripping candidate positions are estimated for workpieces with cylindrical or differently sized portions, then the robot can handle various workpiece shapes, but the machine learning precision degrades due to multiple correct answers for the same input data

Engineering Contradiction:

Improveability to handle various workpiece shapesVSAvoidmachine learning inference precision

Core Design Contradiction:

Adaptability or versatilityVSMeasurement precision

Solution Approach 1:

The patent segments the gripping position estimation into two distinct stages: first estimating multiple candidate gripping positions, then selecting the optimal position from these candidates. This segmentation allows the system to maintain adaptability for various workpiece shapes while improving precision by evaluating multiple options rather than relying on a single ambiguous estimation.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The patent introduces dynamic selection of gripping positions based on robot arm configuration and workpiece characteristics. Instead of using a fixed estimation approach, the system dynamically evaluates multiple candidate positions and selects the most appropriate one, thereby maintaining both versatility across different workpiece shapes and precision in position determination.

Inventive Principle:

Principle #15Dynamics

2Productivity

If a single gripping position is calculated for each workpiece, then the machine learning model operates efficiently, but the gripping accuracy decreases for workpieces with symmetric or ambiguous features

Engineering Contradiction:

Improveprocessing efficiencyVSAvoidgripping position accuracy

Core Design Contradiction:

ProductivityVSManufacturing precision

Solution Approach 1:

The patent performs preliminary estimation of multiple candidate gripping positions before final selection. By pre-calculating multiple potential positions and then selecting the optimal one based on additional criteria, the system maintains processing efficiency while improving gripping accuracy for ambiguous workpiece features.

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The patent incorporates feedback mechanisms where the robot arm's actual configuration and capabilities are considered when selecting from candidate gripping positions. This feedback loop ensures that the selected position is not only geometrically valid but also practically achievable, thereby improving gripping accuracy without significantly impacting processing efficiency.

Inventive Principle:

Principle #23Feedback

3Reliability

If the robot estimates multiple candidate gripping positions and selects the best one, then the gripping success rate improves, but the calculation time increases

Engineering Contradiction:

Improvegripping success rateVSAvoidposition calculation time

Core Design Contradiction:

ReliabilityVSLoss of time

Solution Approach 1:

The patent applies partial action by estimating a limited number of candidate gripping positions rather than exhaustively analyzing all possible positions. This approach achieves high gripping success rate by considering the most promising candidates while avoiding the time cost of exhaustive search, thereby balancing reliability with time efficiency.

Inventive Principle:

Principle #16Partial or excessive action

Data Source

PatentUS20250100153A1Information processing apparatus, information processing method, robot system, method for manufacturing a product, and recording medium

Publication Date: 2025.03.27 CANON KK

AI summary

An information processing apparatus includes an information processing portion configured to perform information processing. The information processing portion is configured to obtain sensing data from a vision sensor having sensed a workpiece, use the sensing data as input data for a learned model, and obtain, on a basis of the learned model, information of a region including a plurality of positions that are possible positions of a first finger portion among at least two finger portions included in a robot in a case of causing the at least two finger positions to perform a holding operation of holding the workpiece.