Abstract
We present a scalable algorithm for learning parametric constraints in highdimensions from safe expert demonstrations. To reduce the ill-posedness of theconstraint recovery problem, our method uses hit-and-run sampling to generatelower cost, and thus unsafe, trajectories. Both safe and unsafe trajectoriesare used to obtain a representation of the unsafe set that is compatible withthe data by solving an integer program in that representation's parameterspace. Our method can either leverage a known parameterization or incrementallygrow a parameterization while remaining consistent with the data, and weprovide theoretical guarantees on the conservativeness of the recovered unsafeset. We evaluate our method on high-dimensional constraints forhigh-dimensional systems by learning constraints for 7-DOF arm, quadrotor, andplanar pushing examples, and show that our method outperforms baselineapproaches.