The MAGICAL Benchmark for Robust Imitation

       Sam Toyer   Rohin Shah     Andrew Critch   Stuart Russell
         Department of Electrical Engineering and Computer Sciences
               University of California, Berkeley
          {sdt,rohinmshah,critch,russell}@berkeley.edu

                       Abstract
     Imitation Learning (IL) algorithms are typically evaluated in the same environment
     that was used to create demonstrations. This rewards precise reproduction of
     demonstrat ...