leap-c

Learning Predictive Control

A framework for integrating optimal control solvers into deep learning pipelines.

Combines model predictive control with reinforcement learning and imitation learning
Built on top of community-standard tools acados and CasADi
Seamless integration with PyTorch for deep learning. The OCP solver is wrapped into a PyTorch module and can be differentiated end-to-end.

leap-c is developed through a collaboration between:

Department of Engineering Cybernetics - Norwegian University of Science and Technology (Prof. Sebastien Gros)
Neurobotics Lab - University of Freiburg (Prof. Joschka Boedeker)
Systems Control and Optimization Laboratory - University of Freiburg (Prof. Moritz Diehl)

Documentation latest build: Jun 20, 2025

Contents