Perception Algorithm Offloading Under Resource Contention
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing solutions fail to effectively manage resource contention and failure probability in concurrently executed perception algorithms on mobile computing devices, leading to reduced performance and increased power consumption.
Innovation Solution
A mobile computing device that estimates contention and failure probability for perception algorithms, offloading algorithms to external resources if the failure threshold is exceeded, while ensuring perception-performance and power-consumption constraints are met.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If perception algorithms are executed on more powerful computing platforms, then perception performance and reliability are improved, but power consumption increases
Solution Approach 1:
The system dynamically adjusts the execution configuration of perception algorithms based on real-time operating conditions. The processor selectively executes algorithms with different configurations (e.g., number of iterations, sampling frequency, detection parameters) depending on the device's power state, movement speed, and environmental factors, allowing optimal balance between reliability and power consumption
Solution Approach 2:
The system changes operational parameters of perception algorithms based on operating conditions. When power consumption needs to be reduced, the system adjusts parameters such as reducing the number of detection iterations, lowering sampling frequency, or simplifying algorithm complexity, thereby maintaining acceptable performance while consuming less power
2Reliability
If multiple perception algorithms are executed concurrently, then perception performance is improved, but resource contention increases
Solution Approach 1:
The system performs preliminary estimation of resource contention levels before executing perception algorithms. By predicting the computational load and resource requirements of concurrent algorithms, the system can pre-configure execution strategies, allocate resources appropriately, and avoid excessive contention that would degrade performance
Solution Approach 2:
The system continuously monitors execution results and resource usage of perception algorithms, using this feedback to adjust the configuration and concurrency of algorithm execution. When resource contention is detected, the system adjusts the number of concurrent algorithms or their configurations to maintain optimal performance
3Measurement precision
If perception algorithms are configured with higher precision, then detection accuracy is improved, but computation time increases
Solution Approach 1:
The system applies partial precision to perception algorithms based on operating conditions. Instead of always using maximum precision, the system selectively applies higher detection accuracy only when necessary (e.g., when the device is stationary or moving slowly), and uses lower precision configurations when the device is moving quickly or power is constrained, thereby reducing overall computation time while maintaining sufficient accuracy
Data Source
AI summary
A mobile computing device (MCD, 100) for controlling concurrent execution of perception algorithms is provided, comprising network interface circuitry and processing circuitry. The MCD is operative to acquire (202) sensor data from one or more sensors, representing a physical environment in which the MCD is located, execute (203) an application which is dependent on one or more perception algorithms processing the sensor data, the perception algorithm(s) being subject to respective perception-performance constraints and a power-consumption constraint, determine (204) operating conditions of the MCD which have an impact on perception performance and power consumption of the perception algorithm(s), estimate (205) a level of contention for a local computing resource on which at least one of the perception algorithms and the application, or at least two of the perception algorithms, can be executed concurrently, estimate (207) a probability of failure for at least one of the perception algorithm(s) based on the estimated level of contention, respective current configurations of the perception algorithm(s), a current configuration of the local computing resource, and the determined operating conditions, and if (208) the estimated probability of failure exceeds a failure threshold, offload (210) execution of the at least one perception algorithm to an external computing resource (130), else (208) execute (231) the at least one perception algorithm on the local computing resource.