Whole-Body Safe Control of Robotic Systems with Koopman Neural Dynamics

2026-03-04

AI Summary

This work addresses the challenge of simultaneously achieving accurate trajectory tracking and safety assurance in real-time control of high-dimensional, strongly nonlinear robotic systems. To this end, the authors propose a data-driven approach that leverages Koopman operator theory to embed the nonlinear dynamics into a linear space via learned Koopman embeddings. For the first time, this framework is tightly integrated with the Safe Set Algorithm (SSA) within a unified quadratic programming formulation, enabling concurrent optimal tracking and safety enforcement without requiring an additional safety filter. By combining Koopman-based neural dynamics, SSA, and data-driven modeling, the method demonstrates high-precision trajectory tracking and effective obstacle avoidance on both the Kinova Gen3 manipulator and the Go2 quadruped robot, exhibiting both feasibility and optimality.

Technology Category

• Intelligent Robots: State Estimation
• Natural Language Processing: Safety and Robustness
• Search and Optimization: Learning to Search

Application Category

• Responsible Web: Machine-in-the-loop, human agency and autonomy
• Security and Privacy: Large-scale security measurements
• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs

Abstract

Controlling robots with strongly nonlinear, high-dimensional dynamics remains challenging, as direct nonlinear optimization with safety constraints is often intractable in real time. The Koopman operator offers a way to represent nonlinear systems linearly in a lifted space, enabling the use of efficient linear control. We propose a data-driven framework that learns a Koopman embedding and operator from data, and integrates the resulting linear model with the Safe Set Algorithm (SSA). This allows the tracking and safety constraints to be solved in a single quadratic program (QP), ensuring feasibility and optimality without a separate safety filter. We validate the method on a Kinova Gen3 manipulator and a Go2 quadruped, showing accurate tracking and obstacle avoidance.

Problem

• whole-body safe control
• nonlinear dynamics
• safety constraints
• real-time control
• high-dimensional systems

Innovation

• Koopman operator
• safe control
• quadratic programming
• data-driven dynamics
• whole-body safety