Learning from demonstration
How can a robot learn stable, reactive behaviour from a single example?
Demonstration–behaviour interfaceRobot learning & control · EPFLSthithpragya
Sthithpragya
Gupta
I study how robots can learn from demonstration, transfer skills, and adapt—with analytical guarantees.
Doctoral researcher at LASA, EPFL, advised by Prof. Aude Billard. My work connects robot learning, geometry, and feedback control.
One skill. Different embodiments.Kinematic Intelligence · Science Robotics, 2026
↗Research at the intersection ofLearning from demonstrationDynamical systemsRobot kinematicsManipulation
01 / Research interestsMaking demonstrated
Making demonstrated
behaviour reusable.
Four connected questions guide my work, from a user’s demonstration to execution across robots and changing conditions.
Cross-robot skill transfer
How can the same skill work on robots with different kinematics?
Behaviour–embodiment interfaceAdaptation & coordination
How can learned behaviours adapt to new states and reach their goals together?
Behaviour–execution interfaceLanguage-guided robotics
How can task context guide the selection and adaptation of robot behaviour?
Context–behaviour interface02 / Selected projects
Ideas, methods, and robots.
From analytical structure to physical experiments. Explore the questions, contributions, and results behind the work.
Kinematic Intelligence
Teach a skill once. Execute it across different robots, with their kinematic constraints built into the control policy.
Kinematics · Stable controlExplore
Adaptation & temporal coordination
Adapt learned behaviours to new position–velocity states and timing requirements, and coordinate them through a shared arrival time.
State adaptation · Dexterous manipulationExplore
Action Contextualization
Use language to interpret task context, select robot actions and tune how those actions are executed.
Language grounding · Task planningExplore
03 / Publications
View Google Scholar ↗Published research.
2026
Demonstrate once, execute on many: Kinematic intelligence for cross-robot skill transfer
Science Robotics · 11(113), eaea1995
Reusing demonstrated workspace behaviour across robot embodiments with explicit kinematic grounding.
BibTeX citation
@article{gupta2026kinematic,
title = {Demonstrate once, execute on many: Kinematic intelligence for cross-robot skill transfer},
author = {Sthithpragya Gupta and Durgesh H. Salunkhe and Aude Billard},
journal = {Science Robotics},
year = {2026},
doi = {10.1126/scirobotics.aea1995}
}2026
Compact One-Shot Modeling of High-Dimensional Demonstrations Using Laplacian Eigenmaps
IEEE Transactions on Robotics · 42, 1468–1484
A compact feedback policy learned from one high-dimensional demonstration, with global asymptotic stability.
BibTeX citation
@article{gupta2026lemon,
title = {Compact One-Shot Modeling of High-Dimensional Demonstrations Using Laplacian Eigenmaps},
author = {Sthithpragya Gupta and A. Nayak and Aude Billard},
journal = {IEEE Transactions on Robotics},
year = {2026},
doi = {10.1109/TRO.2026.3666134}
}2025
Cuspidal Redundant Robots: Classification of Infinitely Many IKS of Special Classes of 7R Robots
IEEE Robotics and Automation Letters · 10(12), 12509–12516
Analytical structure for understanding inverse-kinematic solutions in special classes of redundant manipulators.
BibTeX citation
@article{salunkhe2025cuspidal,
title = {Cuspidal Redundant Robots: Classification of Infinitely Many IKS of Special Classes of 7R Robots},
author = {Durgesh H. Salunkhe and Sthithpragya Gupta and Aude Billard},
journal = {IEEE Robotics and Automation Letters},
year = {2025},
doi = {10.1109/LRA.2025.3623011}
}2024
Action Contextualization: Adaptive Task Planning and Action Tuning Using Large Language Models
IEEE Robotics and Automation Letters · 9(11), 9407–9414
Grounding task context in structured robot plans and qualitative and quantitative action parameters.
BibTeX citation
@article{gupta2024contextualization,
title = {Action Contextualization: Adaptive Task Planning and Action Tuning Using Large Language Models},
author = {Sthithpragya Gupta and Kunpeng Yao and Loïc Niederhauser and Aude Billard},
journal = {IEEE Robotics and Automation Letters},
year = {2024},
doi = {10.1109/LRA.2024.3460408}
}2021
Efficient Configuration Exploration in Inverse Dynamics Acquisition of Robotic Manipulators
IEEE International Conference on Robotics and Automation · ICRA 2021
Configuration exploration for data-efficient acquisition of robot inverse dynamics.
BibTeX citation
@inproceedings{khadivar2021exploration,
title = {Efficient Configuration Exploration in Inverse Dynamics Acquisition of Robotic Manipulators},
author = {F. Khadivar and Sthithpragya Gupta and W. Amanhoud and Aude Billard},
booktitle = {IEEE International Conference on Robotics and Automation},
year = {2021},
doi = {10.1109/ICRA48506.2021.9561587}
}04 / In the news
News & coverage.
Stories about Kinematic Intelligence and the research behind cross-robot skill transfer.
New robotic control software avoids jamming their joints
A closer look at Kinematic Intelligence, robot-specific constraints, and the experiments behind cross-robot skill transfer.
Read articleSmart Robotics Kinematic Control
Coverage of the framework for reusing demonstrated skills across robots with different mechanical designs.
Read articleHow to teach the same skill to different robots
EPFL’s research story on teaching a task once and transferring its behaviours across a multi-robot assembly line.
Read article05 / About meLearning meets
Learning meets
analytical structure.
I am a doctoral researcher in robotics at the Learning Algorithms and Systems Laboratory, EPFL, working with Prof. Aude Billard.
My research asks how a demonstration can become a reusable robot program. I develop structured representations and control methods that connect data-efficient learning with stability, kinematic feasibility, and adaptation.
My work spans mathematical analysis and implementation on physical robots, including serial manipulators and dexterous hands. Earlier projects include robot inverse dynamics and assistive robotic mechanisms.
Learning from demonstrationStable dynamical systemsRobot kinematicsDexterous manipulationTask & motion planningLanguage grounding
Education & research experience
Doctoral research · Robotics, Control and Intelligent SystemsLASA, EPFL · Lausanne, Switzerland
Research AssistantLearning Algorithms and Systems Laboratory, EPFL
M.S. in RoboticsÉcole Centrale de Nantes, France
Universitat Jaume I, Spain
Universitat Jaume I, Spain
Research InternAdvanced Robotics Centre, National University of Singapore
B.Tech. in Mechanical – Production and Industrial EngineeringIndian Institute of Technology Delhi, India