Humanoid robot startup Holiday Robotics announced on the 21st that its research team led the paper “..

Humanoid robot startup Holiday Robotics announced on the 21st that its research team led the paper “FlashSAC: Fast and Stable Off-Policy Reinforcement Learning for High-Dimensional Robot Control” won the Outstanding Paper Award (Outstanding Paper Award) at the world’s most prestigious robotics society (RSS) 2026.

RSS publishes select papers every year in a single-session manner. RSS’ main thesis awards are divided into Outstanding Paper, Outstanding Systems Paper, and Outstanding Student Paper. In Korea, a KAIST research team won the System Papers category in 2023, and this is the first time that it has won the Best Papers Award. In particular, it is significant in that the research team of startups, not universities or large research institutes, led the research.

FlashSAC, introduced in the award-winning paper, is a reinforcement learning algorithm that quickly and reliably learns high-dimensional robots with large joints like humanoids. The algorithm improves the limitations that existing methods had to compromise between learning speed and stability, reducing humanoid walking learning, which took several hours, to several minutes.

The researchers verified the performance of FlashSAC on 10 kinds of simulators and more than 60 tasks. In particular, the control policies learned in the simulation are applied to real humanoid robots to implement walking, rotating, and moving stairs, and demonstrate the effectiveness of the technology in real-world environments.

The results of this study were also applied to the humanoid robot “FRIDAY” being developed by Holiday Robotics. In order to put humanoids into industrial sites, it is important to quickly learn and apply motions suitable for various tasks and environments. FlashSAC can significantly reduce the learning time of robot control policies, streamlining the process of developing new motions and applying them to real robots.

FlashSAC is a technology that is being applied not only to academic research but also to actual humanoid product development, a Holiday Robotics official said. “It is the competitiveness of Holiday Robotics to connect the world’s best research results with the performance of actual robots and the speed of development.”

In this paper, Kim Dong-hoo and Lee Ho-joon from Holiday Robotics participated as the main authors, and I Made Aswin Nahrendra and Min Se-hee were also named as co-authors. In addition, Professor Jan Peters, a world-renowned scholar in the field of reinforcement learning and robotics, Professor TU Darmstadt, Professor Danica Kragic of Sweden, and Professor Joo Jae-gul of KAIST participated as co-authors.

Song Ki-young, CEO of Holiday Robotics, said, “This award is an achievement that has been recognized on the world stage for the originality and research competitiveness of the core robotics technology developed by Holiday Robotics,” adding, “We will continue to quickly connect research results to real products based on the highest level of research talent and technology and prove that humanoids create real value in the industrial field.”