LB4TL

The traditional robustness semantics for Temporal Logics is a recursive combination of min/max operations. In case we utilize this in Neuro-symbolic algorithm in MBRL it may result in failure due to non-differentiability issues. There was a couple of trials in the literature to propose a smooth approximation for this symbolic objective function in a Neuro-symbolic training process. I have provided the most scalable smooth approximation that enables us to apply policy optimization for more complex temporal task. The toolbox for this technique is available from here .