Parareal with a Learned Coarse Model for Robotic Manipulation

摘要: A key component of many robotics model-based planning and control algorithms is physics predictions, that is, forecasting a sequence states given an initial state controls. This process slow major computational bottleneck for algorithms. Parallel-in-time integration methods can help to leverage parallel computing accelerate predictions thus planning. The Parareal algorithm iterates between coarse serial integrator fine integrator. challenge devise model computationally cheap but accurate enough converge quickly. Here, we investigate the use deep neural network as in context robotic manipulation. In simulated experiments using engine Mujoco propagator show learned leads faster convergence than physics-based model. We further allows apply scenarios with multiple objects, where not applicable. Finally, conduct on real robot are close real-world pushing objects. Videos at this https URL.