Wrist-Mounted Camera Grasp Control for Multi-Angle Robotics
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing robotic control systems are limited in their ability to grasp objects from various orientations due to restricted camera views, typically allowing only top-down grasps, which restricts the number of possible grasp poses and is not suitable for grasping objects from the side or bottom.
Innovation Solution
A robotic system equipped with a wrist-mounted camera oriented towards the gripper, utilizing deep reinforcement learning and a double deep Q-network approach to learn grasp poses from images, allowing for grasping objects from any angle by mapping input pixels to output motor commands in Cartesian coordinates.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If a top-down camera view is used, then the system is simple to implement, but the range of grasp poses is limited
Solution Approach 1:
The patent transitions from a single top-down camera view to a multi-view camera system that captures images from multiple dimensions and angles. This includes top-down, side, and bottom views, enabling the robot to grasp objects from various orientations and positions, thereby resolving the contradiction between system simplicity and grasp pose versatility.
2Adaptability or versatility
If multiple camera views are used, then the range of grasp poses increases, but the device complexity increases
Solution Approach 1:
The patent employs a universal camera system where multiple cameras serve dual purposes: capturing images for both object recognition and grasp pose determination. This multi-functional approach allows the system to achieve versatile grasp capabilities without proportionally increasing overall system complexity, as the same hardware infrastructure supports multiple functions.
3Adaptability or versatility
If side or bottom views are added, then object manipulation capability improves, but processing resources increase
Solution Approach 1:
The patent performs preliminary processing of images from multiple views by selecting only the most relevant images for grasp pose determination. Instead of processing all captured images equally, the system pre-identifies and selects optimal images based on object position and orientation, thereby reducing computational resource consumption while maintaining enhanced object manipulation capabilities.
Data Source
AI summary
In at least one embodiment, under the control of a robotic control system, a gripper on a robot is positioned to grasp a 3-dimensional object. In at least one embodiment, the relative position of the object and the gripper is determined, at least in part, by using a camera mounted on the gripper.