SASP

Surrogate-Assisted Stochastic Programming with Collective Perception for VR Task Offloading

Xiaoxue Sun , Hongpeng Wang , Xiuli Shao and Pei-Cheng SongNankai University

Abstract

Smart cities can provide infotainment services such as virtual reality (VR) video for autonomous vehicles, but currently, the quality of service is disrupted by a large number of random factors. The ultra-low latency and quality of service requirements of VR video and games require more effective handling of random factor interference.

From the perspective of automobiles, this paper proposes a framework for obtaining historical data through collective perception to drive stochastic programming, considering the effects of multiple stochastic variables in VR task offloading.

For the first time, a Surrogate-Assisted Stochastic Programming (SASP) based on machine learning models is employed to jointly optimize the task offloading time and cost while controlling the correlation of multiple stochastic variables.

Through simulation and experiments, the proposed surrogate-assisted stochastic programming method can effectively utilize the historical data obtained by collective perception, which is conducive to achieving better results in uncertain environments and improving the reliability and stability of VR task offloading.

SASP_ICRA2024.mp4

Video Presentation

Research Motivation

Current bottleneck

Future Infotainment Demand

As autonomous vehicles become more prevalent, there is an increasing demand for advanced infotainment services that offer immersive experiences like VR videos and games. These services need to deliver high-quality content with minimal latency to meet user expectations, making it crucial to address the challenges of integrating such services effectively.

Limited Onboard Computing Power

Current autonomous vehicles have limited onboard computing power, which restricts their ability to support complex infotainment applications, particularly those requiring substantial computation like VR. This limitation hampers the deployment of diverse and high-quality in-vehicle entertainment services, necessitating offloading solutions and enhanced computational support.

Network Service Instability

Urban network services are often unstable due to fluctuating MEC server resources and variable data transmission quality affected by traffic conditions. This instability impacts the quality of infotainment services by causing interruptions or degraded performance, highlighting the need for robust solutions to manage and mitigate these uncertainties.

Solutions and Innovations

Data-Driven Multi-Stochastic Variable Programming Model

Develop a model that acquires and analyzes stochastic factor features from scene information using collective perception in autonomous vehicles, enhancing the understanding of uncertainty factors.

Task Offloading Model for In-Vehicle VR Services

Construct a multi-stochastic variable programming model specifically for offloading computational tasks related to in-vehicle VR services within a cooperative vehicle infrastructure system.

Surrogate-Assisted Stochastic Programming (SASP)

Implement SASP models to accelerate and improve the performance of multiple stochastic variable programming models, addressing the algorithmic complexity and reducing decision time.

Optimization of Latency and Cost

Jointly optimize the latency and cost associated with VR services in the vehicle-road collaboration scenario, improving overall service quality and efficiency.

Methods and Results