Efficient Stochastic Optimal Control through Approximate Bayesian Input Inference
This work addresses stochastic optimal control problems, offering a principled method for robotics or autonomous systems, though it appears incremental by building on existing inference techniques.
The paper tackled the challenge of optimal control under uncertainty by framing it as an input estimation problem using approximate Bayesian inference, resulting in a solver that outperformed baselines on nonlinear simulated tasks.
Optimal control under uncertainty is a prevailing challenge for many reasons. One of the critical difficulties lies in producing tractable solutions for the underlying stochastic optimization problem. We show how advanced approximate inference techniques can be used to handle the statistical approximations principled and practically by framing the control problem as a problem of input estimation. Analyzing the Gaussian setting, we present an inference-based solver that is effective in stochastic and deterministic settings and was found to be superior to popular baselines on nonlinear simulated tasks. We draw connections that relate this inference formulation to previous approaches for stochastic optimal control and outline several advantages that this inference view brings due to its statistical nature.