Deep Learning Based Caching for Self-Driving Cars
▶ Introduction
 

Recently, the automobile industries have focused on the next stage of autonomous driving, called “self-driving”, where cars will drive themselves without human driver intervention [1]. In addition, according to a study on the incremental time and what activities people will perform if everyone uses self-driving cars, it is estimated that there will be 22 billion of hours for extra media consummation in the US [3]. Without steering wheel and driver’s seat, the self-driving car will have new interior outlook and space that can be used as a lounge place. For traveling people, self-driving cars will be new places for engaging in entertainment. In human-driven cars, drivers choose the infotainment contents. However, in the absence of the driver, the self-driving car should determine itself the infotainment contents that are likely to entertain its passengers and do not violate prohibited and restricted content access. However, the choice of infotainment contents depends on the features of the car’s passengers. Therefore, using AI, the self-driving cars should learn themselves and understand passengers’ features for delivering the appropriate infotainment contents to the passengers.

Caching Infotainment contents styled to passenger features via deep learning in autonomous vehicles will be a new business opportunity for automotive, telecom, e-commerce industries for making a business transaction with passengers, where autonomous vehicles can provide heterogonous Infotainment contents and games on payments. Federated Learning (FL) is a more flexible technique that enables the location-centric distributed learning process at autonomous vehicles and Multi-access Edge Computing servers [4,5]. Global Federated Learning Model (GFLM) for passenger and location-centric deep Infotainment services can be trained and tested at Data Center (DC). GFLM combines Convolutional Neural Network (CNN) and Multilayer Perceptron (MLP). To predict people’s features (such as emotion, age, and gender) based on body/face images, GFLM uses CNN. Furthermore, to predict the infotainment services and contents that meet people’s feature, GFLM uses MLP. Then, the trained and tested models can be deployed at MEC servers in close proximity to the self-driving cars (at the RSUs), where each self-driving car can download the models and uses them for predicting passengers’ features via facial images and identifies the infotainment contents that are appropriate to the passengers’ features.

Figure 1. Overview of Passenger and Location-Centric Deep Infotainment Caching

▶ Research Challenges and Issues

  The challenges related to deep learning based caching for self-driving cars are listed as follows:
  • In human-driven cars, drivers choose the infotainment contents. However, in the absence of the driver, the self-driving car should determine itself the infotainment contents that are likely to entertain its passengers and do not violate prohibited and restricted content access. To achieve this, self-driving cars need to learn passengers' features and have infotainment content styled to passengers' features.
  • Even many researchers focus on infotainment services, infotainment services styled to passenger’s body/facial features (e.g., emotion, facial features, gesture, network traffic, etc.) via centralized and distributed learning in both cloud/Data Center (DC) and autonomous vehicles is new and not tackled in existing researches.
  • Self-driving cars will eventually deliver more heterogeneous infotainment contents such as movies, TV, music, and games as well as recent emerging technologies such as Virtual, Augmented, and Mixed Reality [3]. However, obtaining infotainment contents from the DC can induce high car-DC delay. Therefore, self-driving cars need to be supported by MEC servers by caching infotainment content in close proximity to self-driving cars.
  • Self-driving cars are sensitive to delay. Therefore, to reduce car-DC delay and save backhaul bandwidth, communication and caching resource utilization in the MEC servers and self-driving cars need to be optimized.


  • ▶ References

     
    1. M. Daily, S. Medasani, R. Behringer, and M. Trivedi, “Self-driving cars,” Computer, vol. 50, no. 12, pp. 18?23, 2017.
    2. Frost Sullivan, “Global autonomous driving market outlook (Report, march 2018),” https://info.microsoft.com/rs/157-GQE-382/images/K24A2018%20Frost%20%26%20Sullivan%20-%20Global%20Autonomous%20Driving%20Outlook.pdf , [Online; accessed May. 22, 2019].
    3. G. Jarvis, “Keeping entertained in the autonomous vehicle,” TU-Automotive Detroit, 6-7 Jun. 2018 (Novi, Michigan, United States).
    4. Ndikumana, Anselme, Nguyen H. Tran, and Choong Seon Hong. "Deep Learning Based Caching for Self-Driving Car in Multi-access Edge Computing." arXiv preprint arXiv:1810.01548 (2018).
    5. Ndikumana Anselme, and Choong Seon Hong. "Self-Driving Car Meets Multi-Access Edge Computing for Deep Learning-Based Caching," in proceedings of 2019 International Conference on Information Networking (ICOIN), IEEE, 2019.


    ▶ Achievements

     
    1. Anselme Ndikumana, Nguyen H. Tran, Do Hyeon Kim, Ki Tae Kim, and Choong Seon Hong, "Deep Learning Based Caching for Self-Driving Cars in Multi-access Edge Computing," IEEE Transactions on Intelligent Transportation Systems, DOI:10.1109/TITS.2020.2976572
    2. Ndikumana Anselme, and Choong Seon Hong. "Self-Driving Car Meets Multi-Access Edge Computing for Deep Learning-Based Caching," in proceedings of 2019 International Conference on Information Networking (ICOIN), IEEE, 2019.