Project: #406 Towards Privacy-Preserving Networked Autonomous Mobility: Analysis, Tools Development, and Real-World Evaluation Progress Report - Reporting Period Ending: March 30, 2023 Principal Investigator: Ding Zhao Status: Active Start Date: July 1, 2022 End Date: June 30, 2023 Research Type: Advanced Grant Type: Research Grant Program: FAST Act - Mobility National (2016 - 2022) Grant Cycle: 2022 Mobility21 UTC Progress Report (Last Updated: April 2, 2023, 2:22 p.m.) % Project Completed to Date: 15 % Grant Award Expended: 0 % Match Expended & Document: 0 USDOT Requirements Accomplishments The major goal of this project is to study the privacy issues in autonomous vehicles and robots and to protect the private information of users. Task 1 aims to summarize the privacy issues from literature review. Task 2 aims to analyze the potential privacy issue for autonomous devices. Task 3 aims to investigate the privacy issue of the public dataset. Task 4 aims to study the difference between privacy levels. Task 5 aims to study privacy-preserving methods based on the analysis above. Task 6 aims to deploy the privacy-preserving algorithm to the real-world environment. Task 7 is the summary and output of this work. We have finished the investigation over privacy issues in autonomous vehicles and robots with respect to different privacy levels, namely individual level, group level, proprietary level, and attribute level. Based on the investigation, we locate one major problem as the learning from data while preserving individual level privacy requires large ‘noise’ to protect the signal, which causes the learning process tending to fail. In order to protect privacy under such circumstances, we studied the gradient of reinforcement learning in related tasks and the possibility to enhance the gradients. We found that the gradient is denser and more resilient to perturbation if the corresponding parameter contributes to calculation of more data points. Based on the analysis, we proposed a shared-weight transformer. Primary results show that our method can achieve a higher level of privacy preservation. In the future, we will improve our method and make the training process of large models more resilient to the perturbation introduced by the privacy-preserving algorithm. We will test our method in a robot navigation task, namely active perception, where the robot actively chooses moving trajectories to obtain better perception of the environment. This work will be summarized in a paper. Impacts We have published one of the first comprehensive surveys on autonomous vehicles. We have released the survey on arxiv. We will also host a competition at an international conference on the protection of privacy for autonomous vehicles. Other n/a Outcomes New Partners n/a Issues No changes