Research

A published paper and a patent application.

Paper: Reinforcement Learning for Constraint Satisfaction Game Agents

Amritesh Praveen and Gagan Venkat, May 2026. DOI: 10.5281/zenodo.20076630.

The paper works up from tabular reinforcement-learning methods, through deep Q-networks, to an adversarial PPO-trained Wumpus agent that learns to hunt using memory-based tracking rather than direct sight. The system it describes is the game you can play right here, or, as the paper puts it, just play the paper.

Patent application: Railway Track Monitoring and Repairing Device

Indian patent application 202541017469 A, published 7 March 2025. A track-riding trolley with acoustic defect sensors, a robotic repair link with a motorized cutter, GPS, an EMI sensor, and an AI-based imaging unit that monitors gravel distribution.