CONIX Publication

We address the challenge of controlling an unknown non-linear system without using prior data or training. This work was inspired by the control of highly dynamical cyber-physical systems including drones, cars and bipedal robots. Controller design for these systems usually involves both model-based and model-free approaches in a hierarchical fashion, with low-level model-free controllers dealing with disturbances and hardware specifics and high-level model-based controllers providing performance specifications and objectives. Our research aims to simplify the deployment of low-level controllers allowing engineers to focus on high-level task-specific controller design.