Project Synopsis

This project aims to implement an inverse learning framework for Autonomous Vehicles (AVs) for motion-based anomaly detection by modeling and discovering the target decision-making process. We consider the mobility of AVs as a key example of AI-based action planning and aim to identify suspicious activities taken by intruding nodes, as part of NSF Project Number #2204721.

We made some initial tests on extending the trajectory planning for actuator UAVs that include overhead manipulators. The goal is to develop RL algorithms to achieve a desired tip trajectory for a given base trajectory.

Preliminary results can be found in this paper presented in SwarmNet 2023 Workshop

Research Task 7: Anomaly Detection and Safety Monitoring

The ultimate goal of this project is Anomaly Detection, or identifying Agents’ actions that are not fully aligned with the expected rational behavior obtained by Inverse Learning. This Aspect includes (i) developing a reverse engineering framework that monitors the environment and target’s actions to discover its decision-making strategy, as a baseline, and (ii) identifying deviations from predicted behavior. The challenges include projecting the observer’s perception of the environment to the target’s perspective (seeing the world from the target’s eyes), determining the target’s ultimate goal and reward-generation process (reading the agent’s brain), and including potentially unknown factors in the decision-making strategy. To this end, we develop a set of Network-Level Safety Metrics (NSM) to gauge the overall safety of traffic highways with mixt traffic of regular and self-driving vehicles.

For more information, please read this article

Undergraduate Research

TBD

Project Team

PI: Dr. Abolfazl Razi arazi@clemson.edu

Graduate Students:

• Hazim Alzorgan

Undergraduate Students:

• TBD

Outcomes

The following papers are the outcome of completing this project.

• Alzorgan, Hazim, Abolfazl Razi, and Ata Jahangir Moshayedi. “Actuator trajectory planning for uavs with overhead manipulator using reinforcement learning.” 2023 IEEE 34th Annual International Symposium on Personal, Indoor and Mobile Radio Communications (PIMRC). IEEE, 2023.
• Alzorgan, Hazim, and Abolfazl Razi. “Monte Carlo Beam Search for Actor-Critic Reinforcement Learning in Continuous Control.” arXiv preprint arXiv:2505.09029 (2025).