Kundan Kumar
Research Scientist and GenAI Data Scientist
Hi there!
I am a researcher focused on developing intelligent, secure, and adaptable AI systems for next-generation cyber-physical infrastructure. My work spans deep reinforcement learning (DRL), multi-agent systems, large language models (LLMs), safe and explainable AI, computer vision, and robotics, with real-world applications in smart energy systems, autonomous vehicles, and critical infrastructure.
The core of my Ph.D. research centers on developing physics-informed and safety-critical deep reinforcement learning (DRL) frameworks that embed domain knowledge and system constraints directly into the learning process. By incorporating physical laws, safety boundaries, and system dynamics into policy optimization, I design agents capable of making robust, interpretable, and reliable decisions in dynamic, high-stakes environments. My work addresses challenges such as uncertainty quantification, adversarial resilience, and safe exploration, while enabling agents to generalize across diverse network topologies, environmental conditions, and task distributions through advanced transfer learning and meta-learning techniques.
I also leverage the CARLA simulator for autonomous driving research, combining computer vision, trajectory planning, and policy learning in complex traffic environments. My work integrates vision-based perception models for object detection, semantic segmentation, and sensor fusion, enabling robust situational awareness for autonomous agents. In parallel, I integrate LLM-based reasoning into simulation and control frameworks to support high-level planning, adaptive decision-making, and interactive human-AI collaboration for robotics and safety-critical control.
Beyond autonomous and energy systems, my broader research interests include probabilistic modeling, statistical machine learning, and developing AI systems that are robust, trustworthy, and deployable in real-world complex environments.
Other Research Interests:
- Computer Vision: Visual perception, object detection, semantic segmentation, sensor fusion for autonomous systems.
- Software Systems: Scalable software engineering, simulation framework development, real-time systems integration.
- Statistical Machine Learning: Uncertainty quantification, probabilistic modeling, data-driven inference in dynamic environments.
- Robotics: Learning-based control, adaptive planning, safe human-robot interaction, multi-modal robotic systems.
News
| [Jan 2025] | Our paper on Transfer Learning Enhanced Deep Reinforcement Learning for Volt-Var Control in Smart Grids has been accepted to IEEE PES Grid Edge Technologies Conference & Exposition 2025. |
| [Aug 2024] | Our paper on Workshop for High Performance Computing has been accepted at Pittsburgh Supercomputing Center 2024. |
| [Jul 2024] | Our paper on Bayesian Optimization for Deep Reinforcement Learning in Robust Volt-Var Control has been accepted to IEEE PES General Meeting 2024. |
| [Nov 2023] | Our paper Deep RL-based Volt-VAR Control and Attack Resiliency for DER-Integrated Distribution Grids was accepted to IEEE ISGT 2024. |
| [Aug 2022] | Participated in the Oxford Machine Learning Summer School, completing tracks in MLx Health and MLx Finance. |
| [Apr 2022] | Our paper on Pattern-Based Multivariate Regression using Deep Learning (PBMR-DP) was accepted to ICMLA 2022. |