Autonomous Navigation Compute Offloading for Cloud-Based SLAM
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Autonomous vehicles face challenges in precise navigation due to processor-intensive image analysis for SLAM, which can lead to resource constraints and difficulties in orienting themselves, especially when moving and dealing with noise and inaccuracies.
Innovation Solution
A method that offloads computations from local vehicle processors to cloud-based systems, utilizing bandwidth detection and adaptive offloading to distribute processing resources, allowing for efficient SLAM operations by partitioning tasks between local and cloud-based systems.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If image analysis for SLAM is performed locally on the autonomous vehicle, then navigation capability is maintained, but processing resources are overwhelmed and other necessary tasks cannot be executed
Solution Approach 1:
The patent segments the SLAM processing workload into two parts: local processing for critical navigation functions and cloud-based processing for non-critical computations. This allows the vehicle to maintain navigation capability while offloading intensive image analysis tasks to cloud resources, freeing up local processing capacity for other necessary tasks.
Solution Approach 2:
The patent introduces a communication interface as an intermediary between the local vehicle system and cloud-based processing resources. This intermediary manages the transfer of image data and processing results, enabling the system to leverage remote compute resources without compromising local navigation reliability.
2Measurement precision
If more processing resources are allocated to SLAM operations, then orientation accuracy improves, but available processing resources for other tasks decrease
Solution Approach 1:
The patent extends the processing architecture from a single local dimension to include a remote cloud dimension. By distributing SLAM computations across both local and cloud-based processors, the system can achieve high orientation accuracy without concentrating all processing resources in one location, thus avoiding resource conflicts for other tasks.
3Productivity
If SLAM computations are performed with high intensity, then real-time map creation is achieved, but the vehicle lacks processing resources for other necessary tasks
Solution Approach 1:
The patent implements a dynamic processing architecture where the distribution of SLAM computations between local and cloud resources can be adjusted based on available bandwidth and processing needs. This dynamic allocation allows the system to maintain real-time map creation capability while adapting resource distribution to ensure other necessary tasks can execute reliably.
Data Source
AI summary
Embodiments herein include a method executable by a processor coupled to a memory. The processor is local to a vehicle can operable to determine initial location and direction information associated with the vehicle at an origin of a trip request. The processor receives one or more frames captured while the vehicle is traveling along a navigable route relative to the trip request and estimates an execution time for each of one or more computations respective to an analyzing of the one or more frames. The processor, also, off-loads the one or more computations to processing resources of a cloud-based system that is in communication with the processor of the vehicle in accordance with the corresponding execution times.