KoopmanFlow uses a Koopman-inspired structural bias to decouple global steady-state motions from high-frequency local corrections in robotic control policies.
arXiv · March 17, 2026 · 2603.13781
The Takeaway
Generative policies usually struggle to balance stable overall trajectories with the rapid transients needed for contact or occlusion. This spectral decoupling allows robots to handle sudden visual cues without the temporal smoothing artifacts typical of standard ODE-based generative planners.
From the abstract
Generative Control Policies (GCPs) show immense promise in robotic manipulation but struggle to simultaneously model stable global motions and high-frequency local corrections. While modern architectures extract multi-scale spatial features, their underlying Probability Flow ODEs apply a uniform temporal integration schedule. Compressed to a single step for real-time Receding Horizon Control (RHC), uniform ODE solvers mathematically smooth over sparse, high-frequency transients entangled within