- 1d_region_of_attraction_estimate.ipynb shows how to estimate and learn the region of attraction for a fixed policy.
- basic_dynamic_programming.ipynb does basic dynamic programming with piecewise linear function approximators for the mountain car example.
- reinforcement_learning_pendulum.ipynb does approximate policy iteration in an actor-critic framework with neural networks for the inverted pendulum.
- reinforcement_learning_cartpole.ipynb does the same as above for the cart-pole (i.e., the inverted pendulum on a cart).
- 1d_example.ipynb contains a 1D example including plots of the sets.
- inverted_pendulum.ipynb contains a full neural network example with an inverted pendulum.
- adaptive_safety_verification.ipynb investigates the benefits of an adaptive discretization in identifying safe sets for the inverted pendulum.
- lyapunov_function_learning.ipynb demonstrates how a parameterized Lyapunov candidate for the inverted pendulum can be trained with the machine learning approach in [1].
| [1] | S. M. Richards, F. Berkenkamp, A. Krause, The Lyapunov Neural Network: Adaptive Stability Certification for Safe Learning of Dynamical Systems. Conference on Robot Learning (CoRL), 2018. |