Sim-to-Real Transfer for Humanoid Robots
Sim-to-Real Transfer for Humanoid Robots
Half-Day Workshop at the 2025 International Conference on Humanoid Robots
Room 203, COEX, Seoul, Korea | Thursday, October 2, 13:30-17:30
Embedded Files
This workshop will focus on open challenges in sim-to-real transfer for humanoid robots.
This workshop will focus on open challenges in sim-to-real transfer for humanoid robots.
Humanoid robotics has made rapid advances in dynamic control, largely enabled by progress in sim-to-real transfer. However, successful real-world deployment depends critically on how robots are modeled, how the sim-to-real gap is addressed, and how learning pipelines are designed. This workshop focuses on the specific engineering and algorithmic decisions that make robust sim-to-real transfer possible. Through talks and discussions, we aim to share practical insights and foster collaboration among researchers and industry practitioners tackling this challenge. By sharing lessons from real deployments, this workshop aims to accelerate the development of capable, real-world humanoid robots.
Invited Speakers
Invited Speakers
Pulkit Agrawal
MIT
Sven Behnke
University of Bonn
Marco Hutter
ETH Zurich
Beomjoon Kim
KAIST
Stephan Pleines
NVIDIA
Jonah Siekmann
Agility Robotics
Accepted Contributions
Accepted Contributions
HuB: Learning Extreme Humanoid Balance
SLAC: Simulation-Pretrained Latent Action Space for Whole-body Real-World RL
Learning to Walk in Costume: Adversarial Motion Priors for Aesthetically Constrained Humanoids
Online Learning-Based Inertial Parameter Identification of Unknown Object for Model-Based Control of Wheeled Humanoids
Learning Goal-Directed Object Pushing in Cluttered Scenes With Location-Based Attention
RAMBO: RL-Augmented Model-Based Whole-Body Control for Loco-Manipulation
Robot Trains Robot: Automatic Real-World Policy Adaptation and Learning for Humanoids
Expert-Guided Imitation for Learning Humanoid Loco-Manipulation from Motion Capture
RL-based Giant Swing Motions on the Open-Source Robot MEVITA
Learning Resilent Blind Locomotion Controller for Humanoid Robots using Deep Reinforcement Learning
LiPS: Large-Scale Humanoid Robot Reinforcement Learning with Parallel-Series Structures
Trinity: A Modular Humanoid Robot AI System
Residual Dynamics for Reducing the Simulation-to-Real Gap in Zero-Shot Humanoid Skill Deployment
Too Much of a Good Thing: When sim2real Efforts Impede Policy Learning (And What to Do About It)
Multi-Embodiment Locomotion at Scale with extreme Embodiment Randomization
KLEIYN: A Quadruped Robot Evolved from the Open-Source MEVIUS, Equipped with a Active Waist for Chimney Climbing
Towards a Motion Foundation Model: General-Purpose Whole Body Control through Zero-Shot Sim-to-Real Reinforcement Learning
Schedule
Schedule
Organizers
Organizers
Julian Eßer
Fraunhofer IML
Chenhao Li
ETH Zurich
Gabriel Margolis
MIT
Huazhe Xu
Tsinghua University
Page updated
Google Sites
Report abuse