Autonomous robots are increasingly expected to operate safely in complex, time-varying, and partially unknown environments where traditional control barrier function (CBF)–based model predictive control (MPC) suffers from the need for explicit analytic barrier functions and limited adaptability to high-dimensional or unstructured settings. This paper presents a data-driven framework that parameterizes the barrier function with a neural network and enforces the discrete-time CBF condition through an asymmetric training loss, augmented by a critic–learner strategy and empirical calibration to improve sample efficiency and conservativeness. The learned neural CBF is then seamlessly embedded into an MPC formulation to provide formal safety guarantees while preserving performance objectives. Simulation results on a near-realistic MATLAB environment demonstrate that the proposed Neural-CBF–MPC controller achieves robust, real-time safe navigation and outperforms conventional CBF-MPC baselines in dynamic and uncertain scenarios.
- A Neural-Control Barrier Function for Navigation in Complex Environments
- Zhuopeng Yang - Murdoch UniversityHai Wang - Murdoch UniversityAmirmehdi Yazdani - Murdoch University
- Hai Wang (Editor) - Murdoch University
- Proceedings of the Second International Conference on Advanced Robotics, Control, and Artificial Intelligence, pp.564-572
- Lecture Notes in Networks and Systems
- Springer Nature Singapore; Singapore
- 991005919712907891
- School of Engineering and Energy; Harry Butler Institute
- English
- Book chapter