Takuto Otomo

Papers

Total Citations

H-Index

About

Takuto Otomo is a researcher at the forefront of safety and robustness in robot learning, with a focus on deep reinforcement learning (DRL) and its real-world vulnerabilities. His work critically examines the intersection of adversarial machine learning and robotic control, particularly for legged and autonomous systems. In his highly cited 2022 study, "Adversarial joint attacks on legged robots," Otomo demonstrated that adversarial perturbations to actuators can severely compromise the safety and stability of DRL-trained robots, revealing a critical gap in current training paradigms. He further advanced this line of inquiry in 2025 with "Robustness evaluation of offline reinforcement learning for robot control against action perturbations," where he assessed the resilience of offline RL—a promising approach that learns from static datasets—against real-world disturbances. By systematically exposing how both online and offline RL models fail under adversarial conditions, Otomo has provided foundational insights for developing more secure, deployable robotic systems. His work is essential reading for researchers in robotics, adversarial machine learning, and safe AI, and his findings are increasingly cited as benchmarks for robustness testing in autonomous navigation and manipulation.