Yoshihisa Tsurumine
Papers
10
Total Citations
270
H-Index
6
About
Yoshihisa Tsurumine is a leading researcher in deep reinforcement learning (DRL) for robotic manipulation, with a particular focus on deformable objects like cloth. His major contributions center on developing sample-efficient, real-world applicable DRL algorithms that can learn complex control policies directly from high-dimensional raw image inputs, bypassing the need for engineered state representations. His seminal 2018 work on "Deep reinforcement learning with smooth policy update" for robotic cloth manipulation has garnered 179 citations, establishing a foundational approach for stable policy learning in this challenging domain. Tsurumine has pioneered the integration of generative adversarial imitation learning (GAIL) with DRL to overcome the costly burden of manual reward function design, as demonstrated in his 2019 paper (24 citations) and subsequent 2022 work on goal-aware imitation from imperfect demonstrations. He has also advanced sim-to-real transfer through innovative techniques like cyclic policy distillation (2023, 16 citations) and domain randomization, enabling robots to deploy learned policies in the real world with minimal human effort. His notable achievements include implementing DRL on power-efficient FPGA hardware for edge robotics (2021) and extending multi-agent reinforcement learning to cooperative grasping tasks (2025).
Research Focus
Key Achievements
Top Papers
- 1
- 2
- 3
- 4
- 5Deep dynamic policy programming for robot control with raw images14 citations · 2017
- 6
- 7
- 8
- 9
- 10