Karol Arndt
Papers
7
Total Citations
81
H-Index
4
About
Karol Arndt is a robotics researcher specializing in sim-to-real transfer, reinforcement learning, and adaptive robot learning — fields that address one of the most fundamental challenges in modern robotics: bridging the gap between simulation and real-world deployment. His most influential contribution, DROPO (2023, 33 citations), introduced a principled offline domain randomization method that enables more efficient and accurate transfer of learned policies from simulation to physical robots, advancing the state of the art in dynamics parameter optimization. Complementing this work, his research on meta reinforcement learning for sim-to-real adaptation (2020, 17 citations) demonstrated how policies can be trained to rapidly adapt across domains, reducing reliance on costly real-world data collection. Arndt has also made notable strides in safe robot learning through SafeAPT (2022, 11 citations), which leverages diverse simulated policies to minimize safety risks during real-world deployment. His earlier work on affordance learning for visuomotor control (2019, 11 citations) highlighted his interest in modular, data-efficient neural architectures for end-to-end robot control. More recently, his exploration of co-imitation pushes boundaries into simultaneous robot design and behavior learning, reflecting a broad and forward-looking research vision.
Research Focus
Key Achievements
Top Papers
- 1DROPO: Sim-to-real transfer with offline domain randomization33 citations · 2023
- 2Meta Reinforcement Learning for Sim-to-real Domain Adaptation17 citations · 2020
- 3
- 4Affordance Learning for End-to-End Visuomotor Robot Control11 citations · 2019
- 5Co-imitation: Learning Design and Behaviour by Imitation4 citations · 2023
- 6Online vs. Offline Adaptive Domain Randomization Benchmark3 citations · 2023
- 7Few-Shot Model-Based Adaptation in Noisy Conditions2 citations · 2021