Disturbance Injection Under Partial Automation: Robust Imitation Learning for Long-Horizon Tasks

Partial Automation (PA) with intelligent support systems has been introduced in industrial machinery and advanced automobiles to reduce the burden of long hours human operation. Under PA, operators perform manual operations (providing actions) that switch automatic/manual mode (mode-switching). Since PA reduces total duration operation, these two action mode-switching can be replicated by imitation learning high sample efficiency. To this end, letter proposes Disturbance Injection under (DIPA) as a novel framework. In DIPA, actions (in mode) are assumed observables each state used learn both policies. The above is robustified injecting disturbances into operator's optimize disturbance's level for minimizing covariate shift PA. We experimentally validated effectiveness our method long-horizon tasks simulations real robot environment confirmed outperformed previous methods reduced demonstration burden.