Predictive SLAM Using Prior Session Data for AR Headsets
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Previous head-mounted display systems for augmented and virtual reality require tethering to external computing devices due to high processing demands for real-time simultaneous localization and mapping (SLAM), limiting mobility and efficiency in determining positional state information.
Innovation Solution
The implementation of a predictive SLAM system that uses pre-captured and stored images of landmarks to enhance localization accuracy without the need for brute force methods, allowing the head-mounted display to operate independently by comparing real-time SLAM frames with previously mapped trajectories and data from a SLAM repository.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If real-time SLAM processing is performed using brute force methods, then localization accuracy can be achieved, but processing requirements and device complexity increase significantly
Solution Approach 1:
The system performs preliminary SLAM processing during off-line periods to generate candidate trajectories and store them in a data structure. This pre-computation reduces the complexity of real-time processing by providing pre-prepared localization data that can be quickly retrieved and refined during active use, thereby maintaining accuracy while reducing processing demands.
Solution Approach 2:
The SLAM processing is divided into off-line batch processing and real-time incremental processing. The off-line phase handles computationally intensive tasks of generating candidate trajectories, while the real-time phase focuses on refining these trajectories and updating positional state. This segmentation allows complex processing to be distributed across different time periods, reducing peak processing requirements.
2Power
If tethering to external computing devices is required, then processing capabilities are sufficient, but mobility and ease of operation are limited
Solution Approach 1:
The system extracts and stores essential SLAM data structures and candidate trajectories in local memory during off-line processing. This allows the head-mounted display to operate independently without continuous tethering to external computing devices, as the critical processing data is already available locally for real-time refinement and use.
Solution Approach 2:
Computationally intensive SLAM processing is performed in advance during off-line periods, preparing candidate trajectories and localization data before the user needs them. This pre-computation enables the device to function autonomously during real-time operation, eliminating the need for continuous external tethering while maintaining processing capabilities.
3Productivity
If continuous tethering to external systems is used, then processing demands are met, but mobility and user freedom are reduced
Solution Approach 1:
The system alternates between off-line batch processing modes and real-time incremental processing modes. During off-line periods, comprehensive SLAM processing is performed to build and refine candidate trajectories. During real-time use, the system performs lighter incremental updates. This periodic processing strategy maintains productivity while eliminating continuous tethering requirements.
Solution Approach 2:
The system performs preliminary and comprehensive SLAM processing during off-line periods, preparing all necessary candidate trajectories and localization data in advance. This allows the head-mounted display to operate freely during real-time use without continuous external connection, as the processing groundwork has already been laid.
Data Source
AI summary
An information handling system operating a predictive simultaneous localization and mapping (SLAM) system may comprise a camera mounted to a wearable headset capturing an image of a landmark, a positional sensor mounted to the wearable headset measuring a current positional state of the headset, and a network adapter receiving a stored SLAM frame, including an identification and position of the landmark within a previously captured image. A processor executing code instructions of the predictive SLAM system may generate a current SLAM frame based on the image captured via the camera, determine a Kalman gain associated with an observed location of the landmark based on the current SLAM frame and the stored SLAM frame, and determine a corrected position of the wearable headset based on a previously measured positional state of the headset, the current positional state of the headset, the current SLAM frame, and the Kalman gain.