Timezone: »

Sample Efficient Path Integral Control under Uncertainty
Yunpeng Pan · Evangelos Theodorou · Michail Kontitsis

Thu Dec 10 08:00 AM -- 12:00 PM (PST) @ 210 C #42

We present a data-driven stochastic optimal control framework that is derived using the path integral (PI) control approach. We find iterative control laws analytically without a priori policy parameterization based on probabilistic representation of the learned dynamics model. The proposed algorithm operates in a forward-backward sweep manner which differentiate it from other PI-related methods that perform forward sampling to find open-loop optimal controls. Our method uses significantly less sampled data to find analytic control laws compared to other approaches within the PI control family that rely on extensive sampling from given dynamics models or trials on physical systems in a model-free fashion. In addition, the learned controllers can be generalized to new tasks without re-sampling based on the compositionality theory for the linearly-solvable optimal control framework.We provide experimental results on three different systems and comparisons with state-of-the-art model-based methods to demonstrate the efficiency and generalizability of the proposed framework.

Author Information

Yunpeng Pan (Georgia Institute of Technolog)
Evangelos Theodorou (Georgia Tech)
Michail Kontitsis (Georgia Institute of Technology)

More from the Same Authors