Dibyendu Das

I am a doctoral candidate in the Computer Science department of Stony Brook University, New York.

Interests

Artificial Intelligence  •  Deep Reinforcement Learning  •  Cloud & Data Infrastructure  •  Distributed Systems  •  Stochastic Decision Making  •  Robotics

Projects

• Self Evaluation: Robot evauating its confidence in performing complex manipulation tasks [code] Paper: [arXiv,2024] Self-evaluation is a probabilistic algorithm that enables the Robot to evaluate whether, for a particular task to be performed in a given workspace, it has a sufficient number of demonstrations in order to generate feasible manipulation plans while satisfying the task-related motion constraints. The underlying method relies upon an \(\epsilon\)-optimal arm identification in a Multi-Armed Bandit setup using \((\epsilon,\delta)\)-PAC (Probably Approximately Correct ) learning.
• Ultimate Tic-Tac-Toe: 9 Synchronized Tic-Tac-Toe Boards [code] An AI agent inspired by the AlphaZero algorithm and trained using deep Reinforcement Learning techniques such as Monte-Carlo Tree Search and Deep Q-Network , for playing the non-trivial board game of Ultimate Tic-Tac-Toe, where 9 small Tic-Tac-Toe boards work in tandem, giving rise to a gameplay more complex than Othello . The agent — trained using self-play against itself for 7 days on a low-end GPU — defeated an average human player with Elo rating of 1400+ in chess 8 out of 10 times.
• UnreLyzer: A static interval analyzer for C like unreliable programs [code] Paper: [arXiv,2016] Unrelyzer is a static analysis tool that analyzes a program defined by a subset of C (Mini-C) grammar with an addendum where each arithmetic, boolean and memory (Read/Write) operation in the program is probabilistic/unreliable in nature. This tool statically analyzes the program using Abstract Interpretation in the interval domain and determines the interval of values with a certain confidence for each program variable at each program point. This tool was built as part of my Master’s thesis.

Publications

• Screw Geometry Meets Bandits: Incremental Acquisition of Demonstrations to Generate Manipulation Plans, in IEEE International Conference on Robotics and Automation (ICRA), Vienna, Austria, 2026. Dibyendu Das, Aditya Patankar, Nilanjan Chakraborty, C.R. Ramakrishnan, and I.V. Ramakrishnan

• Transferring Kinesthetic Demonstrations across Diverse Objects for Manipulation Planning, in IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), Hangzhou, China, 2025. Dibyendu Das, Aditya Patankar, Nilanjan Chakraborty, C.R. Ramakrishnan, and I.V. Ramakrishnan

• Knowledge-Enabled Motion Generation for Complex Manipulation Tasks, in IEEE ICRA Workshop on Geometric Representations: The Roles of Modern Screw Theory, Lie Algebra, and Geometric Algebra in Robotics, London, UK, 2023. Dibyendu Das, Aditya Patankar, Fumi Honda, Dasharadhan Mahalingam, Nilanjan Chakraborty, C.R. Ramakrishnan, and I.V. Ramakrishnan

• Select or Suggest? Reinforcement Learning-based Method for High-Accuracy Target Selection on Touchscreens, in Proceedings of the 2022 CHI Conference on Human Factors in Computing Systems, pp. 1-15. Zhi Li, Maozheng Zhao, Dibyendu Das, Hang Zhao, Yan Ma, Wanyu Liu, Michel Beaudouin-Lafon, Fusheng Wang, I.V. Ramakrishnan, and Xiaojun Bi

• Work In Progress: Verifying Stability Guarantees of Control Software Implementations in the Presence of Sensor Level Faults, in Proceedings of the 13th ACM International Conference on Embedded Software (EMSOFT). Saurav Kumar Ghosh, Debasmita Lohar, Dibyendu Das, and Soumyajit Dey

• Failure Estimation of Behavioral Specifications, in International Symposium on Dependable Software Engineering: Theories, Tools, and Applications (SETTA), pp. 315-322, Springer International Publishing. Debasmita Lohar, Anudeep Dunaboyina, Dibyendu Das, and Soumyajit Dey