Hierarchical Task and Motion Planning using Logic-Geometric Programming ( HLGP )

In this work, we present a hierarchical approach to task and motion planning (TAMP) within an optimizationbased framework. Recent work on formulating TAMP as a logic enhanced nonlinear program has shown remarkable capabilities. However, scaling this approach to domains with many discrete decisions or longer horizons implies a computational bottleneck. To overcome this, we introduce hierarchies within this framework, where on coarser levels a problem with less discrete decisions is solved. Formally, the hierarchies are defined in a way that the resulting nonlinear programs on coarser hierarchy levels are lower bounds on the finer hierarchies. We demonstrate the generality of the approach for both a bi-manual manipulation task and a mobile manipulation scenario which includes a “worm” like walking robot.