Sim Home 🏠
Why simulators?
- training policies with reinforcement-learning methods
- identifying system parameters via gradient-based regression
- generating datasets for learning
- differentiable model representations in model-predictive-control frameworks
- general Monte Carlo testing and validation
- avoiding damaging physical platforms by prototyping in sim
Where to go from here?
Differentiable Simulators
Sim Assumptions
Sim Comparison Table
Subtasks | Formulation | Assumptions
Computing Dynamics (Lagrangian Formulation, Newton-Euler)
Computing Contact (Soft, Hard) (NCP, LCP)
Maximal vs. Minimal Coordinates
Actuator Models
Sensor Models
Obtaining Gradients
Parameterizing Damping/LPM coefficients