My spotlight talk at IROS 2021 Workshop on Safe Real-World Robot Autonomy is online and summarizes some of the more recent work on predictive safety filters.
Check out our new preprint on Learning-based Moving Horizon Estimation through Differentiable Convex Optimization Layers, considering the case of partial state measurements in a constrained learning-based control setting.
Our recently published article called Nonlinear learning-based model predictive control supporting state and input dependent model uncertainty estimates provides a modular framework to incorporate learning-based prediction models in different MPC applications such as set point stabilization and economic system operation.
We’ve proposed a method in our latest article to improve the performance of a model predictive controller using available system trajectories with constraint violations. The underlying mechanism is based on a soft constraint formulation, which supports polytopic terminal sets.
In our new preprint, we propose a mechanism to recover infeasible nonlinear model predictive control problems in an asymptotically stable fashion.
We designed and implemented a predictive safety filter for a miniature racing application and combined it with imitation learning to safely learn an expert policy: Preprint
The Article “A predictive safety filter for learning-based control of constrained nonlinear dynamical systems” has been accepted as a regular paper in Automatica. An updated version of the preprint can be found here.
We’ve published a preprint, which provides a theoretical analysis of ‘Bayesian MPC’ in terms of performance and expected safety during learning: `Performance and safety of Bayesian model predictive control: Scalable model-based RL with guarantees‘
I’m a phd student at the intelligent control systems group at IDSC ETH. My research interests are in the fields of Safe learning, Model Predictive Control, Bayesian Optimization, and Autonomous Driving, see also my Google Scholar profile.
For a detailed CV, see my LinkedIn profile.