Nitin Desai

Nitin Desai is an Assistant Professor at the Bajaj Engineering Skill Training (BEST) Centre, Mahindra University, where he works on robot learning and intelligent automation for Industry 4.0. He has submitted his doctoral thesis in Robotics at the Indian Institute of Technology Kharagpur, as a Prime Minister’s Research Fellow under Prof. Sankha Deb and Prof. Debashis Sen. He holds an M.Tech. in CAD/CAM and Automation from VJTI Mumbai and a B.E. in Mechanical Engineering from DKTE’s Textile and Engineering Institute, ranking third in both.

His doctoral work took on a difficult form of the manipulation problem: picking textureless, occluded parts out of cluttered bins, and running robots where people should not be, beside a steelmaking furnace. It covers the full perception-to-action loop, from occlusion-resilient 6-DoF pose estimation and learned grasp prediction to an underactuated three-finger hand and collision-free planning in deep bins. He has since moved toward language-conditioned manipulation with LLMs and vision-language models, reinforcement learning for adaptive control, and generative synthetic data for sim-to-real transfer. He has worked with Scale AI on post-training data for large language models. His work appears in ASME MSEC, ASME IMECE, IEEE INDICON, Intelligent Service Robotics and Mechanics Based Design of Structures and Machines.

• 08/2021 – 03/2026: Ph.D. in Robotics (thesis submitted), Advanced Technology Development Centre, Indian Institute of Technology Kharagpur, India
  • Dissertation/Thesis Title: Deep Learning Based Vision-Guided Robotic Bin-Picking, Adaptive Multi-Fingered Grasping, and Autonomous Metallurgical Sampling for Challenging Environments
  • Supervisors: Prof. Sankha Deb (Mechanical Engineering) and Prof. Debashis Sen (Electronics and Electrical Communication Engineering)
• 06/2019 – 08/2021: M.Tech. in CAD/CAM and Automation, Veermata Jijabai Technological Institute (VJTI), Mumbai, India
  • Dissertation/Thesis Title: Parametric Design of Fasteners Using FreeCAD
• 08/2014 – 06/2017: B.E. in Mechanical Engineering, DKTE Society’s Textile and Engineering Institute, Ichalkaranji (Kolhapur), India
  • Dissertation/Thesis Title: Vibration Analysis of Band Saw Machine

Journal Articles

• N. Desai, S. Deb, D. Sen, “Occlusion-resilient pose estimation of textureless components in a cluttered environment and its implementation in robotic bin picking,” Intelligent Service Robotics, Springer, 2026, DOI: 10.1007/s11370-026-00711-8.
• N. Desai, S. Ansary, A. Mishra, S. Deb, D. Sen, “Design and implementation of a lance for performing robot-assisted metallurgical temperature measurement and sampling and robot trajectory planning,” Mechanics Based Design of Structures and Machines, Taylor & Francis, 2025, DOI: 10.1080/15397734.2025.2579754.

Conference Papers

• N. Desai, “Enhancing autonomy and adaptive control of industrial robots in dynamic environments using reinforcement learning,” ASME International Mechanical Engineering Congress & Exposition (IMECE 2025), Hyderabad, 2025.
• N. Desai, G. Behera, S. Deb, D. Sen, “Development of a three-finger adaptive robotic gripper for handling irregular and complex objects and its dynamic grasp synthesis,” Proc. ASME 20th International Manufacturing Science and Engineering Conference (MSEC 2025), Greenville, SC, USA, 2025, DOI: 10.1115/MSEC2025-155617.
• N. Desai, G. Behera, S. Deb, D. Sen, “A genetic algorithm based adaptive and hybrid pathfinding algorithm to enhance the performance of robotic bin-picking,” 2024 IEEE 21st India Council International Conference (INDICON), 2024, DOI: 10.1109/INDICON63790.2024.10958511.
• G. Behera, N. Desai, “Deep learning-based automatic machining feature recognition from CAD model,” ASME International Mechanical Engineering Congress & Exposition (IMECE 2025), Hyderabad, 2025.
• G. Behera, N. Desai, S. Deb, “Implementation of soft computing-based metaheuristic algorithms in multi-objective environmentally-conscious machining operation sequence optimization with carbon emission reduction,” Proc. ASME 20th International Manufacturing Science and Engineering Conference (MSEC 2025), Greenville, SC, USA, 2025, DOI: 10.1115/MSEC2025-155708.
• N. Desai, S. Ansary, A. Mishra, S. Deb, D. Sen, “Design and implementation of a lance for performing robot-assisted metallurgical temperature measurement and sampling and robot trajectory planning and control,” International Conference on Next Generation Technologies: Design and Manufacturing (ICNGT 2024), IIT Madras, 2024.
• N. Desai, S. Deb, D. Sen, “MaskPose: 3D pose estimation and image segmentation of objects from RGB images for robotic grasping,” National Seminar on Innovation & Technology Inclusion Program in Manufacturing and Asset Management for Steel Industry, Tata Steel Ltd., Jamshedpur, 2023.

• 03/2022 – 11/2025: Teaching Assistant (Computer Integrated Manufacturing; Non-Traditional Manufacturing), Department of Mechanical Engineering, Indian Institute of Technology Kharagpur
• 04/2025 – 07/2025: Technical Consultant (LLM post-training data and evaluation), Scale AI (remote)
• 03/2022 – 05/2025: Research Scholar, Centre of Excellence in Advanced Manufacturing Technology, Indian Institute of Technology Kharagpur (multi-sensor robotic cell for hazardous steelmaking operations, with Tata Steel)
• 08/2022 – 05/2025: Teaching Assistant (PMRF), Mechatronics and Robotics, Government ITI Hijli, Kharagpur
• 06/2020 – 06/2021: Intern, Centre for Machine Intelligence and Data Science, Indian Institute of Technology Bombay
• 06/2020 – 06/2021: Teaching Assistant (Robotics; Finite Element Analysis), Department of Mechanical Engineering, VJTI Mumbai

Research areas: Robot learning; 6-DoF perception and grasp synthesis in clutter; language-conditioned manipulation with LLMs and VLMs; reinforcement learning for sequential decision-making; generative models and sim-to-real transfer; underactuated hand design; motion planning; field robotics for hazardous manufacturing.

Factories are one of the few places where robots already work at scale, yet most industrial arms still replay fixed programs inside carefully engineered cells. Desai’s research asks what it takes for a robot to perceive, reason and act reliably once that structure is removed: parts arrive jumbled in bins, surfaces are textureless and occluded, instructions come in plain language, and the work site is too hazardous for people. His approach combines learned perception, model-based planning and purpose-built hardware, and he validates on physical industrial robots, not in simulation alone.

Perception in clutter. Occlusion-resilient 6-DoF pose estimation of textureless industrial parts, instance segmentation and pose recovery from RGB images (MaskPose), and CNN-based grasp affordance heatmaps reaching 0.97 IoU on complex geometries, deployed for vision-guided bin picking. Language and foundation models for manipulation. Grounding spoken and written instructions in visual scene understanding with LLMs and VLMs, so that a command such as “pick the cylinder” becomes a feasible pick-and-place plan executed through inverse kinematics. He is building agentic pipelines that couple perception, reasoning and control for autonomous pick-and-place.

Learning for sequential decision-making. Deep Q-learning for multi-step assembly sequence planning and reinforcement learning for adaptive control of industrial manipulators in non-stationary environments, alongside evolutionary and sampling-based planners (GA-based hybrid planning, RRT) for collision-free motion in confined bins. Generative models and sim-to-real. GAN-based and simulation-driven (Isaac Sim, RoboDK) synthetic data to cover the long tail of grasping and placement scenarios and cut the cost of real-world data collection. Mechanism design and field robotics. An underactuated three-finger adaptive hand that passively switches between precision and power grasps, with 95% grasp success across metallic, plastic and ceramic parts, designed with a physics-based simulator for contact, slip and force closure; a robot-mounted lance for molten-metal temperature measurement and sampling, developed with Tata Steel; and a self-centering in-pipe inspection robot with CNN-based defect detection.

Going forward, he is interested in robot foundation models that adapt to a new factory task from a handful of demonstrations, and in making learned manipulation policies predictable enough to be trusted on the shop floor.