Probabilistic optimal control

Probabilistic Optimal Control:
Path integrals, linear Bellman equations and estimation-control dualities

Organizers: Bert Kappen and Emanuel Todorov

Optimal control theory concerns the problem of how to act optimally when reward is only obtained at a later time, and is central to modeling intelligent behavior in animals and machines. Examples are control of multi-joint robot arms, navigation of vehicles, coordination of multi-agent systems. In addition, optimal control theory plays an important role in financial applications.

While optimal control in its general form remains hard, recent developments show that the problem can be substantially simplified if certain structure is imposed. The simplifications include linearity of the (Hamilton-Jacobi) Bellman equation and duality with Bayesian estimation and allow for analytical computation of the optimal control laws and closed form expressions of the optimal value functions [1, 2]. These theoretical developments have given rise to new algorithms which resemble probabilistic inference more than traditional control, and have proven remarkably effective in several applications.

The aim of the workshop is to provide an overview of this nascent field. The program will consist of one or two introductory talks, followed by presentations on novel developments.

[1] H. J. Kappen, A linear theory for control of non-linear stochastic systems, Physical Review Letters 95:200201, 2005.
[2] E. Todorov, Linearly-solvable Markov decision problems, Advances in Neural Information Processing Systems 19, 1369-1376 2007.