Possible Project
Project-1: External Status Display for Autonomous Vehicles - This project explores how an autonomous vehicle can communicate its current status and driving intentions to surrounding road users through an external display. The system may show information captured by sensors such as vehicle speed, driving direction, turning intention, acceleration or deceleration, and potential warning states, helping nearby drivers, cyclists, or pedestrians better understand and anticipate the vehicle’s behavior. (reference)
Project-2: Failure-Aware Autonomous Vehicles - This project investigates how autonomous vehicles should communicate failures or unexpected behaviors to nearby users. Students will design structured prompts and use generative AI to create video scenarios in which an autonomous vehicles (such as car, ground and aerial drone) experience problems such as stopping unexpectedly, becoming blocked, or rerouting. Different communication strategies, such as no explanation, warning signals, or short status messages, can then be compared through a user study to evaluate their effects on perceived safety, comfort, and predictability. (reference)
Project-3: LLM-Based Explanations for Autonomous Vehicle Decisions - This project investigates whether large language models can generate high-quality, human-like explanations for autonomous vehicle decisions. Using datasets such as BDD-X, which provide driving scenarios, vehicle actions, and human-written explanations, students will develop a methodology that provides the driving context and vehicle decision to an LLM and asks it to generate a concise explanation of why the action was taken. The generated explanations can then be compared with human annotations in terms of correctness, relevance, clarity, completeness, and hallucination, with the goal of identifying prompting or reasoning strategies that enable LLMs to produce explanations approaching the quality of human-generated ones. (reference)
Project-4: Adaptive Sensor Fusion under Sensor Degradation - This project investigates how autonomous vehicles can maintain reliable perception when individual sensors become degraded or temporarily fail. Using multi source data such as camera and LiDAR from a public autonomous-driving dataset, Students will use datasets containing adverse sensing conditions and sensor corruptions, and may additionally simulate controlled failures when necessary. A lightweight AI-based reliability estimator will then assess the quality of each sensor and dynamically adjust their contribution to the fusion process. The project will compare single-sensor, fixed-fusion, and adaptive-fusion approaches to determine whether reliability-aware sensor fusion can provide more robust perception under imperfect sensing conditions. (reference)
Project-5: Context-Aware Smart-Ring Interaction for Autonomous Drones - This project investigates how wearable sensing can provide intuitive high-level control of autonomous indoor drones. Building on smart-ring gesture interaction, students will use IMU-based finger movements to recognize commands such as moving, stopping, approaching, or inspecting an area, while additionally considering contextual information such as the drone's current task, state, and relative position. By combining gesture and context information, the system aims to infer user intent more accurately and reduce ambiguity between similar movements. Students will compare gesture-only and context-aware approaches and evaluate their recognition accuracy, responsiveness, and usability without requiring low-level drone flight programming.