Research
Our overarching goal is to secure decentralized systems and design intelligent autonomous agents that can operate reliably in adversarial, resource-constrained, and safety-critical environments. We approach this by combining blockchain security analysis, machine learning–driven cybersecurity, and reinforcement learning for autonomous UAVs.
Our research sits at the intersection of systems security and artificial intelligence. On one hand, we investigate emerging vulnerabilities in blockchain Layer-2 systems and propose defense frameworks grounded in game theory and AI. On the other, we design intelligent UAV navigation, energy harvesting, and intrusion detection systems that adapt under hostile or uncertain conditions.
Here are some themes and techniques that we currently work on:
Blockchain Layer-2 Security.
Public blockchains like Ethereum provide strong security guarantees but suffer from
scalability limitations, which has led to the rise of Layer-2 (L2) solutions such as
optimistic rollups and payment channel networks. While these systems improve throughput
and reduce transaction costs, they introduce new and underexplored security
vulnerabilities. Our research systematically investigates these weaknesses, including
RPC-based state manipulation attacks, arbitrage exploits in ERC-721 token transactions,
Sybil attacks in shard-based systems, and fake hashed key attacks on payment channels.
To address these threats, we design defense mechanisms that combine graph-based AI
models for anomaly detection, Bayesian game theory for strategic validator behavior, and
cryptographic safeguards to ensure fairness and reliability. Through this line of
work—highlighted by our systems xPoz-Hub, OptiGame, RollGuard, PAROLE, and
FAKey—we aim to make L2 blockchains not only scalable but also secure, dependable,
and economically robust, paving the way for their safe adoption in decentralized
finance, supply chains, and other critical infrastructures.
AI-Enabled UAV Path Planning and Security.
Autonomous UAVs face adversarial challenges such as hostile defense systems and cyber
intrusions. We design federated deep reinforcement learning (FedRL) and economic RL
methods to enable resilient trajectory planning for individual UAVs and swarms. In
parallel, we develop lightweight intrusion detection systems (IDSs) using neuro-fuzzy
inference systems (ANFIS) and PWM-signal analysis, enabling secure UAV navigation with
low computational overhead.
Energy Harvesting for UAVs and IoT Systems. Sustainable autonomous systems require self-sufficient energy solutions. We propose RIS-assisted UAVs and cell-free massive MIMO deployments to enable RF energy harvesting for IoT devices. These methods couple advanced communications theory with deep learning optimization to extend operational lifetime in UAV and CPS deployments.
Cyber-Physical Systems (CPS) Security.
We investigate data-driven anomaly detection and false data injection attacks in CPS
domains such as smart grids and digital healthcare systems. Our approaches combine
bio-inspired optimization, deep learning classifiers, and blockchain-based auditing
mechanisms. By bridging control theory, AI, and blockchain, our work provides resilience
in critical infrastructure systems.
Federated & Privacy-Aware Learning. A cross-cutting theme is the design of federated AI systems that respect privacy and operate securely across distributed, adversarial environments. Our provisional patent on federated DRL-assisted UAV trajectory planning highlights the potential of privacy-preserving collaborative learning in mission-critical applications.
