Robot Grasp Strategy Selection Using ML and Sensor Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Robots face challenges in determining the appropriate manner to grasp objects due to the complexity of selecting the correct grasping strategy, which is difficult for them to replicate like humans.
Innovation Solution
The use of machine learning models to process sensor data, such as vision data, to select a grasp strategy from a plurality of candidate strategies, influencing end effector poses, grasp directions, and manipulation parameters, allowing the robot to determine the optimal grasp approach and interaction with objects.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If traditional programming methods are used to control robot grasping, then the control logic is simple and transparent, but the robot cannot adapt to diverse objects and complex manipulation tasks
Solution Approach 1:
The patent replaces traditional mechanical control systems with machine learning models that process sensor data and generate grasp strategies. The ML models substitute for complex if-then logic programming, enabling the robot to learn optimal grasping strategies from data rather than being explicitly programmed for each scenario.
Solution Approach 2:
The system changes the control parameters from fixed programmed values to dynamic predictions generated by machine learning models. The grasp strategy, end effector pose, and manipulation parameters are determined by ML models based on real-time sensor inputs, allowing continuous adaptation to different objects and situations.
2Adaptability or versatility
If machine learning models are used to determine grasp strategies, then the robot can adapt to diverse objects and tasks, but the computational complexity and processing requirements increase
Solution Approach 1:
The patent segments the grasping task into multiple independent machine learning models, each responsible for specific aspects: object detection, grasp strategy selection, end effector pose determination, and manipulation planning. This modular approach allows each model to be optimized independently and reduces the computational burden on any single model.
Solution Approach 2:
The system performs preliminary processing of sensor data to extract relevant features before feeding them to the grasp strategy model. Pre-computed object properties and pre-defined grasp strategy templates are used to reduce the search space and accelerate real-time decision-making.
3Reliability
If multiple candidate grasp strategies are evaluated, then the quality of grasp selection improves, but the time required to determine the optimal strategy increases
Solution Approach 1:
The patent uses machine learning models to directly predict the optimal grasp strategy parameters from sensor data, rather than evaluating multiple candidate strategies through complex simulations. The ML models have been trained to output high-quality grasp strategies directly, eliminating the need for time-consuming evaluation of numerous alternatives.
Solution Approach 2:
The system uses pre-trained machine learning models that have learned optimal grasp strategies from large datasets of successful grasps. These models copy proven strategies from training data and adapt them to new situations, avoiding the need to evaluate and discover strategies from scratch for each new object.
4Measurement precision
If the end effector pose is precisely determined based on grasp strategy and sensor data, then the grasp accuracy improves, but the computational processing time increases
Solution Approach 1:
The patent replaces complex geometric and kinematic calculations with machine learning models that directly predict end effector pose from sensor data. The ML models have learned the mapping from object features to optimal poses during training, enabling rapid pose determination without extensive real-time computation.
Solution Approach 2:
The system pre-computes and stores relationships between object features and optimal end effector poses during the training phase. During operation, the model quickly retrieves and adapts pre-computed pose information based on real-time sensor inputs, significantly reducing online computation time while maintaining high precision.
Data Source
AI summary
Grasping of an object, by an end effector of a robot, based on a grasp strategy that is selected using one or more machine learning models. The grasp strategy utilized for a given grasp is one of a plurality of candidate grasp strategies. Each candidate grasp strategy defines a different group of one or more values that influence performance of a grasp attempt in a manner that is unique relative to the other grasp strategies. For example, value(s) of a grasp strategy can define a grasp direction for grasping the object (e.g., “top”, “side”), a grasp type for grasping the object (e.g., “pinch”, “power”), grasp force applied in grasping the object, pre-grasp manipulations to be performed on the object, and/or post-grasp manipulations to be performed on the object.