A brief analysis of the technical research on autonomous driving cache

Once self-driving cars become a reality, passengers will no longer worry about safety issues and will need to find new ways to entertain themselves. However, the high latency of car-to-data center (DC) communication will hinder the retrieval of entertainment content from content delivery services. This paper proposes a deep learning-based cache for self-driving cars by using a deep learning method deployed on a multi-access edge computing (MEC) structure. Through simulation tests, it is found that this method can minimize latency.

Recently, autonomous driving technology has made significant progress, and some companies, such as Google, Uber, Samsung, Tesla, Mercedes-Benz, Baidu, etc., have begun to focus on the next stage of autonomous driving - "driverless driving", that is, cars can drive autonomously without human driving intervention. In addition, in order to make autonomous vehicles more intelligent, cars need to be equipped with smart sensors and analysis tools to collect and analyze heterogeneous data related to on-board occupants, pedestrians, and the environment in real time, in which deep learning plays an important role.

The importance of multi-access edge computing in the future

Even though autonomous vehicles have an on-board unit (OBU) that handles computation, communication, caching, and control (4C), the autonomous vehicle resources for 4C are still limited and require assistance from a remote cloud. For effective autonomous vehicle data analytics, low latency and reliable computing are required. However, the reliance on the cloud may affect the performance of autonomous vehicle data analytics due to the associated end-to-end latency. Therefore, to reduce the end-to-end latency, we consider multi-access edge computing (MEC) as a technology suitable for supporting autonomous vehicles for edge analytics. MEC was recently introduced by the European Telecommunications Standards Institute (ETSI) to complement cloud computing, where MEC servers are deployed at the network edge of 4C. In this work, MEC servers are deployed on RoadSide Units (RSUs) for edge analytics and content caching near autonomous vehicles.

With deep learning and 4C features in self-driving cars, passengers will no longer be limited to in-car radio and TV, but will spend more time watching media, playing games, and social networking. However, retrieving these contents from a data center (DC) will make content delivery services worse due to the associated end-to-end latency and consumed backhaul bandwidth resources. As an example, watching a video in a car requires three components, namely the video source, the screen, and the sound system. Therefore, if the video source is not in the car, the car needs to download it from the DC. Assuming that the DC is located far away, the in-car service will experience high latency, and caching in self-driving cars will play an important role in improving the user experience.

The Challenge of Caching in Autonomous Vehicles

For people traveling, autonomous vehicles will become a new entertainment venue. Therefore, content providers and game developers need to seize this new opportunity by providing high-quality entertainment content. However, there is still a lack of literature on how to perform caching of entertainment content in autonomous driving.

Autonomous vehicles can provide more heterogeneous entertainment content such as movies, TV, music, and games as well as recently emerging platforms such as virtual reality (VR). However, the 4C resources of autonomous vehicles are limited. Therefore, MEC servers need to support autonomous vehicles.

Autonomous vehicles are sensitive to latency. Therefore, reducing car-DC latency and saving backhaul bandwidth requires strengthening and optimizing communication and cache resource utilization in MEC servers and autonomous vehicles.

There has been no good answer to how to solve the caching problem in self-driving cars.

Autonomous driving cache based on deep learning

To address the above challenges, Anselme Ndikumana et al. from the Department of Computer Science and Engineering at Kyung Hee University in South Korea proposed using deep learning-based caching and the 4C approach in MEC to improve entertainment services in autonomous vehicles.

Their main methods are summarized as follows:

Passengers have different content preferences, and their choices depend on age and gender. In order to meet the needs of passengers in self-driving cars, a convolutional neural network (CNN) approach is used to predict their age and gender through facial recognition. Specifically, the CNN output is used by the self-driving car in order to determine which entertainment content (e.g., music, videos, and game data) is suitable for the passengers and therefore needs to be cached.

Providing passengers with appropriate entertainment content requires MEC and DC to support autonomous vehicles. In DC, they proposed a MultiLayer Perceptron (MLP) framework to predict the probability of requesting content in a specific area of ​​the autonomous vehicle. The MLP prediction output is then deployed at the MEC server (RSU) in close proximity to the autonomous vehicle. During off-peak hours, each MEC server uses the MLP output for downloading and then caches content with a high probability of request. MLP was chosen over other prediction methods such as AutoRegressive (AR) and Autoregressive Moving Average (ARMA) models, and MLP has the ability to handle both linear and nonlinear prediction problems.

For content that needs to be cached, the autonomous vehicle needs to download the MLP output from the MEC server and then compare it with the CNN output. For comparison, this method also combines k-means and binary classification.

Using deep learning with the 4C components in MEC, a cache for entertainment services is formulated in autonomous vehicles to minimize content download latency.

As autonomous driving technology is not yet fully mature, the technology of autonomous driving caching studied in this paper is still far away, but the methods and research ideas proposed in it are still worth learning from. According to the investigation, this is the first article to study entertainment content caching for autonomous driving cars, in which the caching decision is based on MLP, CNN and available communication, caching and computing resources.