4.5.23-AI-Fan

Conference Video|Duration: 31:02
April 5, 2023
Please login to view this video.
  • Video details
    Learning-enabled data-driven methods have demonstrated impressive empirical performance on challenging autonomous systems. But this performance comes at the cost of reduced transparency and lack of guarantees on the safety or stability of the systems. In this talk, I will present several of our recent efforts that combine machine learning with formal methods and control theory to enable the design of dependable and safe autonomous systems. The approach we took, called neural certificates, provides supervision during training by allowing safety and stability requirements to influence the training process. As a result, the learned policies can achieve a much-improved performance on safety and stability, especially on complex autonomous systems with a large number of agents, following nonlinear and nonholonomic dynamics and needing to satisfy high-level specifications.
Locked Interactive transcript
Please login to view this video.
  • Video details
    Learning-enabled data-driven methods have demonstrated impressive empirical performance on challenging autonomous systems. But this performance comes at the cost of reduced transparency and lack of guarantees on the safety or stability of the systems. In this talk, I will present several of our recent efforts that combine machine learning with formal methods and control theory to enable the design of dependable and safe autonomous systems. The approach we took, called neural certificates, provides supervision during training by allowing safety and stability requirements to influence the training process. As a result, the learned policies can achieve a much-improved performance on safety and stability, especially on complex autonomous systems with a large number of agents, following nonlinear and nonholonomic dynamics and needing to satisfy high-level specifications.
Locked Interactive transcript