Dongheui Lee

We present a scalable framework for cross-embodiment humanoid robot control by learning a shared latent representation that unifies motion across humans and diverse humanoid platforms, including single-arm, dual-arm, and legged humanoid robots. Our method proceeds in two stages. First, we construct a decoupled latent space that captures localized motion patterns across different body parts using contrastive learning, enabling accurate and flexible motion retargeting even across robots with diverse morphologies. To enhance alignment among embodiments, we introduce tailored similarity metrics that combine joint rotation and end-effector (EE) positioning for critical segments, such as arms. Then, we train a goal-conditioned control policy directly within this latent space using only human data. Leveraging a conditional variational autoencoder (c-VAE), our policy learns to predict latent space displacements guided by intended goal directions. We show that the trained policy can be directly deployed on multiple robots without any adaptation. Furthermore, our method supports the efficient addition of new robots to the latent space by learning only a lightweight robot-specific embedding layer. The learned latent policies can also be directly applied to the new robots. Experimental results demonstrate that our approach enables robust, scalable, and embodiment-agnostic robot control across a wide range of humanoid platforms.

We propose a control framework that integrates model-based bipedal locomotion with residual reinforcement learning (RL) to achieve robust and adaptive walking in the presence of real-world uncertainties. Our approach leverages a model-based controller—comprising a Divergent Component of Motion (DCM) trajectory planner and a whole-body controller—as a reliable base policy. To address the uncertainties of inaccurate dynamics modeling and sensor noise, we introduce a residual policy trained through RL with domain randomization. Crucially, we employ a model-based oracle policy, which has privileged access to ground-truth dynamics during training, to supervise the residual policy via a novel supervised loss. This supervision enables the policy to efficiently learn corrective behaviors that compensate for unmodeled effects without extensive reward shaping. Our method demonstrates improved robustness and generalization across a range of randomized conditions, offering a scalable solution for sim-to-real transfer in bipedal locomotion.

The increasing presence of robots in human workspaces underscores the need for intelligent systems that can understand human behaviors and act accordingly for a natural human-robot interaction (HRI). In this work, we propose a method to generate a robot’s behavior for social HRI by integrating both human and robot intentions into the robot’s decision-making process. Our system learns appropriate robot behaviors in social scenarios by observing human-human interactions (HHI). Using a transformer-based model, we first capture the dynamics of each individual and then iteratively adapt both human and robot behavior to achieve a successful interaction. By connecting our model with a human-to-robot motion retargeting framework, our system learns how a robot should behave solely from observing human data. To address the disparity between HHI and HRI, we employ several loss functions that encourage our robot to reproduce the social dynamics observed in humans. As a result, our approach outperforms the state-of-the-art in dyadic human motion forecasting prediction in the largest dataset available and obtains high-quality robot behaviors in human-robot interaction scenarios. Finally, we conduct a thorough evaluation of our performance for HHI, and HRI, and implement and test the system in the real-world TIAGo++ robot.

Understanding human collaborative behavior in tasks with physical interaction is essential for advancing physical human-robot collaboration. Investigating how individuals learn to collaborate over repeated interactions can provide valuable insights for developing robotic agents capable of gradually improving coordination and collaboration performance. Therefore, this study investigated learning behavior in a high-precision task over repeated haptic collaboration. Specifically, we examined if learned collaboration behavior is partner-specific, what collaboration strategies are developed, and if interpersonal differences affect collaboration. Our results indicate that repeated physical collaboration with the same partner allowed for immediate high performance with a familiar partner in subsequent collaborations, whereas adapting to an unfamiliar partner required retraining. Participants used partner-specific collaboration behaviors—in terms of motions and forces—that could be retained in subsequent interactions. Collaborators reduced the variability of their arm motions over repeated collaboration, achieving higher performance, likely due to increased predictability. Collaboration also enabled knowledge transfer between partners, with individual improvement being enhanced when paired with a better-performing partner. These findings suggest that partners in a collaborative precision task optimize their performance by gradually negotiating a joint action strategy, which is reused in subsequent collaborations with familiar partners and carries over to solo task execution.

The concept of priority has been introduced to the robotic systems in 1980s as an effort to overcome problems caused by singularity and it has attracted considerable attention from both the robotics and control societies. However, none of the previous works have successfully addressed two fundamental degenerate properties of singularity: nonsmoothness and imperfect inversion. This technical note proposes a prioritized output tracking control method that guarantees the ultimate boundedness and the convergence of higher priority outputs when the input–output feedback linearization becomes nonsmooth and imperfect by singularity. For that purpose, we first prioritize the input–output feedback linearization and discuss how two degenerate properties can occur when the system becomes singular. Then, we introduce the differential inclusion to deal with the nonsmooth internal dynamics and establish a condition using the upper Dini derivative for asymptotically stable zero dynamics. Also, we propose a strictly positive realness-based feedback gain design and find a condition related to an M-matrix in order to handle nonlinearities that appear in the imperfect feedback linearization. Finally, we combine these two results and show the ultimate boundedness and the output convergence. In addition, we provide a motivational example with a planar four-link manipulator to illustrate the effectiveness and limitations of the proposed method.

Humanoid robots have the potential to mimic human motions with high visual fidelity, yet translating these motions into practical physical execution remains a significant challenge. Existing techniques in the graphics community often prioritize visual fidelity over physics-based feasibility, posing a significant challenge for deploying bipedal systems in practical applications. This article addresses these issues through bounded residual reinforcement learning (RL) to produce physics-based high-quality motion imitation onto legged humanoid robots that enhance motion resemblance while successfully following the reference human trajectory. Our framework, Imitation to Control Humanoid Robots Through Bounded Residual Reinforcement Learning (I-CTRL), reformulates motion imitation as a constrained refinement over nonphysics-based retargeted motions. I-CTRL excels in motion imitation with simple and unique rewards that generalize across five robots. Moreover, our framework introduces an automatic priority scheduler (APS) to manage large-scale motion datasets when efficiently training a unified RL policy across diverse motions. The proposed approach signifies a crucial step forward in advancing the control of bipedal robots, emphasizing the importance of aligning visual and physical realism for successful motion imitation.