Max Robotics — Global Service Robot Directory + US Certification

About

Hiroshi Kera is a researcher at the forefront of safe and resilient robotic control, specializing in the intersection of deep reinforcement learning, adversarial robustness, and fault-tolerant systems. His work critically addresses the vulnerabilities of legged robots—particularly quadrupeds—to real-world perturbations, including adversarial joint attacks and actuator failures. In his highly cited 2022 study on adversarial joint attacks, Kera demonstrated how subtle perturbations to actuator commands can significantly compromise the safety and stability of reinforcement learning-trained robots, highlighting a critical security gap in embodied AI. His 2021 work introduced an adaptive curriculum dynamics randomization (ACDR) algorithm, a pioneering method that enables quadruped robots to maintain control even after actuator failure, a vital capability for operations in remote or extreme environments. More recently, Kera has extended his robustness evaluations to offline reinforcement learning, assessing how action perturbations affect robot control when learning occurs solely from static datasets. With a combined citation count of over 12 for his most impactful papers, Kera’s contributions are shaping the next generation of dependable, failure-resistant autonomous systems.