Koopman DCM: Unstable Eigenfunctions as Data-Driven Representations for Legged Balancing
This paper generalizes Divergent Components of Motion as Koopman eigenfunctions, learning unstable eigenpairs from only 1 hour of real robot data. On a real biped, the learned Koopman DCMs improve reference walking pattern tracking and provide state-based viability constraints when combined with model predictive control.
Paper · arXiv:2607.18760Background
In legged locomotion, Divergent Components of Motion (DCMs) have emerged as characteristic states for balance control by isolating the unstable mode of the dynamics. However, existing DCM formulations apply only to reduced models such as the linear inverted pendulum. Traditional Koopman analysis targets eigenvalues near zero, capturing conserved or slowly varying quantities, and has not been applied to actively search for unstable eigenpairs for control applications.
Core Innovation
This work proposes a novel perspective: generalizing DCMs as Koopman eigenfunctions. Unlike classical Koopman analysis, the authors deliberately search for unstable eigenpairs with large eigenvalues, yielding data-driven DCMs that can be trained exclusively from real robot data without requiring an accurate dynamics model. The learned Koopman DCMs serve dual purposes: improving gait tracking performance and providing state-based viability constraints when integrated with model predictive control (MPC).
Key Results
On a real biped robot, Koopman DCMs learned from only one hour of robot data improve tracking of reference walking patterns. The authors further demonstrate that the learned DCMs provide effective state-based viability constraints when combined with MPC, enabling the controller to proactively avoid unsafe states. Experimental validation on real hardware confirms the practical feasibility of the approach.
Limitations
Validation is currently limited to a single biped platform, and generalization across different morphologies (full-size humanoids, quadrupeds) remains to be demonstrated. While one hour of data is remarkably efficient, performance under extreme perturbations or on unknown terrain has not been tested. The linearization error inherent to Koopman theory may grow in highly nonlinear regimes.
Industry Implications
This approach offers a model-free data-driven path to balance control for legged robots, of significant relevance to the rapidly advancing humanoid robotics industry. The ability to learn stable balance strategies from modest amounts of real-world data could accelerate deployment of bipedal and humanoid robots in logistics, manufacturing, and service applications requiring dynamic locomotion capabilities.