Robot Grasp Training With Flexible Approach Direction Prediction

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing robotic gripping systems face limitations in flexibility regarding approach direction, particularly for parallel grippers, which restricts their application to specific gripper types and orientations.

Innovation Solution

A method for training a machine learning model that predicts gripping contact points and orientations by using a mixture distribution of spherical distributions, allowing for flexible approach directions and overcoming limitations in reachability and visibility constraints.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Device complexity

If a fixed approach direction is used for gripping, then the control complexity is reduced, but the adaptability to different objects and environments is limited

Engineering Contradiction:

Improvecontrol complexityVSAvoidadaptability to different objects

Core Design Contradiction:

Device complexityVSAdaptability or versatility

Solution Approach 1:

The patent applies dynamics by making the approach direction adaptive rather than fixed. The machine learning model dynamically determines the approach direction based on object characteristics, allowing the system to adapt to different objects while maintaining manageable control complexity through automated decision-making.

Inventive Principle:

Principle #15Dynamics

Solution Approach 2:

The patent changes the parameter of approach direction from a fixed value to a variable determined by the machine learning model. This allows the system to optimize the approach direction based on object properties, resolving the contradiction between control simplicity and adaptability.

Inventive Principle:

Principle #35Parameter changes

2Ease of operation

If the robot is restricted to specific gripper orientations, then the reachability is improved, but the versatility of gripping different objects is reduced

Engineering Contradiction:

ImprovereachabilityVSAvoidversatility of gripping

Core Design Contradiction:

Ease of operationVSAdaptability or versatility

Solution Approach 1:

The patent makes the gripper orientation dynamic by using the machine learning model to determine optimal orientations based on object characteristics. This resolves the contradiction by allowing the system to achieve both reachability (through constrained optimization) and versatility (through adaptive selection).

Inventive Principle:

Principle #15Dynamics

Solution Approach 2:

The patent performs preliminary analysis of object characteristics using the machine learning model to determine feasible gripper orientations before execution. This allows the system to pre-compute optimal orientations that satisfy both reachability constraints and versatility requirements.

Inventive Principle:

Principle #10Preliminary action

3Measurement precision

If only visible object parts are considered for gripping, then the measurement accuracy is improved, but the ability to grip objects with occluded parts is reduced

Engineering Contradiction:

Improvemeasurement accuracyVSAvoidability to grip occluded objects

Core Design Contradiction:

Measurement precisionVSAdaptability or versatility

Solution Approach 1:

The patent performs preliminary 3D reconstruction and analysis of object geometry before gripping execution. This allows the system to infer the positions of occluded contact points based on visible parts and object models, resolving the contradiction between measurement accuracy and the ability to handle occluded objects.

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The patent introduces 3D object models and geometric reasoning as intermediaries between visible sensor data and occluded contact points. This mediator allows the system to accurately infer hidden geometry and determine gripping points on occluded surfaces without direct observation.

Inventive Principle:

Principle #24Intermediary (Mediator)

4Device complexity

If a single gripper type is used, then the device complexity is reduced, but the adaptability to different gripper requirements is limited

Engineering Contradiction:

Improvedevice complexityVSAvoidadaptability to gripper requirements

Core Design Contradiction:

Device complexityVSAdaptability or versatility

Solution Approach 1:

The patent implements universality by designing a machine learning-based control system that can adapt to multiple gripper types and requirements. The system determines optimal approach directions and contact points that are applicable to various gripper configurations, making a single system versatile enough to handle different gripper requirements without increasing physical device complexity.

Inventive Principle:

Principle #6Universality (Multi-functionality)

Data Source

PatentUS20250037025A1Method for training a machine learning model for controlling a robot to manipulate an object

Publication Date: 2025.01.30 ROBERT BOSCH GMBH

AI summary

A method for training a machine learning model for controlling a robot. The method includes, for each training data element of a set of training data elements, wherein each training data element comprises training input information about the location of surface points of a respective object and one or more possible approach directions of the robot for manipulating the object, ascertaining, via the machine learning model, one or more contact points, ascertaining, via the machine learning model, weighting parameter values of a mixture distribution of spherical distributions for the approach direction, and training the machine learning model to reduce a loss that contains an approach-direction loss component per training data element and per possible approach direction.