Object-Focused 3D Reconstruction With Sampling-Based Camera Planning

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current systems for autonomous robots to interact with objects in unknown environments face computational intensity and inefficiency in 3D reconstruction and motion planning, particularly in determining accurate camera trajectories for object grasping.

Innovation Solution

The implementation of a system-on-a-chip (SOC) with a deep convolutional network (DCN) for 3D model reconstruction, which includes a general-purpose processor, neural processing units, and multiple processing blocks to compute depth maps, construct object maps, and determine optimal camera trajectories using sampling-based planning and motion planning graphs.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If traditional computational methods are used for 3D reconstruction and motion planning, then measurement precision and reliability are improved, but productivity and use of energy deteriorate due to computational intensity

Engineering Contradiction:
Improve3D reconstruction accuracyVSAvoidcomputational efficiency
Core Design Contradiction:
Measurement precisionVSProductivity

Solution Approach 1:

The patent replaces traditional mechanical computational methods with a neural network-based system. The neural network is trained offline to learn the mapping from 2D images to 3D structures and motion plans, enabling real-time inference without intensive online computation. This substitution of mechanical computation with learned models resolves the contradiction between precision and productivity.

Inventive Principle:
Principle #28Mechanics substitution (Replace mechanical system)

Solution Approach 2:

The system performs preliminary training of the neural network offline before actual operation. During runtime, the pre-trained network quickly processes images to generate 3D reconstructions and motion plans. This preliminary action transfers computational burden from runtime to training time, improving real-time productivity while maintaining accuracy.

Inventive Principle:
Principle #10Preliminary action

2Manufacturing precision

If detailed 3D reconstruction and motion planning are performed, then object grasping accuracy is improved, but use of energy and computational burden increase

Engineering Contradiction:
Improveobject grasping accuracyVSAvoidcomputational energy consumption
Core Design Contradiction:
Manufacturing precisionVSUse of energy by moving object

Solution Approach 1:

The patent substitutes energy-intensive traditional computational algorithms with a neural network that has learned efficient representations during training. The network performs rapid inference to generate grasping plans, significantly reducing computational energy consumption while maintaining or improving grasping accuracy through learned patterns.

Inventive Principle:
Principle #28Mechanics substitution (Replace mechanical system)

Solution Approach 2:

The system changes the computational parameters by using a pre-trained neural network with fixed weights for inference, rather than performing iterative optimization at runtime. This parameter change from optimization-based computation to direct network evaluation reduces energy consumption while preserving the ability to generate accurate grasping plans.

Inventive Principle:
Principle #35Parameter changes

3Reliability

If complete environment scanning and analysis are performed before motion planning, then reliability of object interaction is improved, but loss of time and productivity worsen

Engineering Contradiction:
Improveobject interaction reliabilityVSAvoidplanning time
Core Design Contradiction:
ReliabilityVSLoss of time

Solution Approach 1:

The patent extracts the computationally intensive planning and reconstruction tasks from real-time operation and performs them offline during network training. The trained network then quickly processes incoming images to generate motion plans, extracting only the essential processing needed at runtime. This extraction resolves the contradiction by eliminating time-consuming operations from the operational timeline.

Inventive Principle:
Principle #2Taking out (Extraction)

Solution Approach 2:

The system performs preliminary analysis and learning during the offline training phase, where the neural network learns to interpret images and generate motion plans from training data. During actual operation, this preliminary learning enables rapid decision-making without requiring complete re-analysis of the environment, significantly reducing planning time while maintaining interaction reliability.

Inventive Principle:
Principle #10Preliminary action

Data Source

PatentEP3405845B1Object-focused active three-dimensional reconstruction
Publication Date: 2023.10.25 QUALCOMM INC
  • EP3405845B1 patent drawingFigure 1
  • EP3405845B1 patent drawingFigure 2
  • EP3405845B1 patent drawingFigure 3A

AI summary

A method for guiding a robot equipped with a camera to facilitate three-dimensional (3D) reconstruction through sampling based planning includes recognizing and localizing an object in a two-dimensional (2D) image. The method also includes computing 3D depth maps for the localized object. A 3D object map is constructed from the depth maps. A sampling based structure is grown around the 3D object map and a cost is assigned to each edge of the sampling based structure. The sampling based structure may be searched to determine a lowest cost sequence of edges that may, in turn be used to guide the robot.