Papers
7
Total Citations
161
H-Index
6
About
Rituraj Kaushik is a robotics and machine learning researcher whose work sits at the intersection of data-efficient reinforcement learning, sim-to-real transfer, and adaptive robot control. His research addresses one of the field's most pressing challenges: enabling robots to learn effective behaviors with minimal real-world data — a critical constraint given the cost and risk of physical experimentation. Kaushik's most influential contribution, "Black-box Data-efficient Policy Search for Robotics" (2017, 93 citations), advanced model-based reinforcement learning by leveraging uncertain dynamical models to dramatically reduce the data requirements for robot policy learning. Building on this foundation, he has explored repertoire-based online adaptation, developing methods that allow robots to rapidly recover from damage or unexpected conditions by drawing on diverse pre-learned policy libraries. His work on SafeAPT further addresses the sim-to-real gap by enabling safe policy transfer with safety guarantees during real-world deployment. More recently, Kaushik has extended his research into meta-learning for fast online adaptation and variable impedance control, pushing toward human-like compliant manipulation. Across his career, his publications have accumulated over 160 citations, reflecting meaningful influence on data-efficient and adaptive robot learning — topics increasingly vital as robotics moves into unstructured, real-world environments.
Research Focus
Key Achievements
Top Papers
- 1Black-box data-efficient policy search for robotics93 citations · 2017
- 2Adaptive Prior Selection for Repertoire-Based Online Adaptation in Robotics31 citations · 2020
- 3
- 4
- 5
- 6Data-Efficient Reinforcement Learning for Variable Impedance Control6 citations · 2024
- 7Black-Box Data-efficient Policy Search for Robotics2 citations · 2017