Adaptive AV Compute Offloading for Energy-Latency Balance
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Autonomous vehicles face limitations in processing power due to limited space and energy constraints, which affects their computational efficiency and driving range, as more resource-intensive computations increase energy consumption and reduce battery life.
Innovation Solution
An adaptive architecture that dynamically offloads computing tasks between on-board and off-board systems, prioritizing resource-intensive computations for the cloud and less intensive ones for the vehicle, optimizing processing load based on environmental demands and computational resource usage.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If resource-intensive computations are performed on-board the autonomous vehicle, then computational accuracy and processing capability are improved, but energy consumption increases and battery life decreases
Solution Approach 1:
The patent segments computational tasks into different categories based on their resource requirements and urgency. Complex, non-time-critical computations (e.g., semantic map creation, fleet-wide data analysis) are separated and executed in the cloud, while time-critical computations (e.g., obstacle detection, immediate navigation decisions) remain on-board. This segmentation allows the system to achieve high computational accuracy for critical functions while minimizing overall energy consumption by offloading non-critical heavy computations.
Solution Approach 2:
The patent introduces a cloud computing system as an intermediary to handle resource-intensive computations. The on-board system communicates with the cloud system, which acts as an external computational resource. This intermediary approach allows the autonomous vehicle to access powerful computational resources without carrying them physically, thereby maintaining high computational capability while reducing on-board energy consumption and extending battery life.
2Use of energy by moving object
If more computing subsystems are off-loaded to the cloud, then on-board energy consumption is reduced, but computational latency and response time may increase
Solution Approach 1:
The patent applies local quality by differentiating between time-critical and non-time-critical computational tasks. Time-critical functions such as obstacle detection, collision avoidance, and immediate navigation adjustments are kept on-board with high processing priority to ensure minimal latency. Non-time-critical functions such as semantic map creation, route optimization, and fleet data analysis are off-loaded to the cloud. This differentiated approach ensures that energy is conserved on-board while critical response times are maintained.
Solution Approach 2:
The patent implements dynamic task allocation where the boundary between on-board and cloud computations is not fixed but adapts based on real-time conditions. The system dynamically adjusts which computations are performed locally versus remotely based on factors such as network availability, computational urgency, and on-board energy levels. This dynamic approach allows the system to optimize the balance between energy consumption and response time adaptively, rather than using a static division of computational tasks.
Data Source
AI summary
A method may include obtaining sensor data relating to an autonomous vehicle (AV) and a total measurable world around the AV. The method may include identifying an operating environment of the AV and determining a projected computational load for computing subsystems that facilitate a driving operation performable by the AV corresponding to the identified environment. The method may include off-loading first computing subsystems of the computing subsystems in which computations of the first computing subsystems may be processed by an off-board cloud computing system and processing computations associated with second computing subsystems of the computing subsystems by an on-board computing system. The method may include obtaining first computational results corresponding to the computations processed relating to the first computing subsystems and determining the driving operation of the AV based on the first computational results and second computational results corresponding to computations processed relating to the second computing subsystems.