We present an algorithm for optimal guidance of users in road networks. It is a "stochastic-on-time-arrival (SOTA)"-like algorithm which calculates optimal guidance strategies with reliable paths, for road network origin-destination pairs. Our contribution consists here in extending an existing SOTA algorithm, in order to include robustness of the guidance strategy, towards path failures. The idea of SOTA algorithms is to calculate the maximum probability of reaching a destination node, starting from any node of a road network, and given a time budget. This calculus gives the optimal path for every origin destination pair of nodes in the network, with an associated optimal adaptive guidance strategy, and with respect to the considered time budget. The approach models very well the importance of the travel time variability in the route choice process. Indeed, by adopting this kind of guidance strategies, users may accept short deteriorations on the average travel time, if in return, they have a guarantee on the reliability of the travel time (in term of the maximum probability of reaching their destinations in given time budgets). We propose here an extension of this approach in order to take into account the existence and the performance of alternative detours of the selected paths, in the calculus of the guidance strategy. We take into account the fact that one or many links of the selected optimal path may fail during the travel. We then consider that users may be sensitive to path changing. That is to say that they may prefer paths with efficient alternative detours, with respect to paths without, or with less efficient detours, even with a loss in the average travel time, and/or in its reliability. In order to take into account such behaviors, we propose a model that includes the existence as well as the performance of detours for selected paths, in the calculus of the travel time reliability (i.e. the maximum probability of reaching a destination node). This new way of calculating travel time reliability guarantees a kind of robustness of the guidance strategies. That is to say that, the travel time reliability associated to the obtained optimal guidance strategy is not likely to change, however associated adaptive paths change during the travel. The variation of the travel time reliability, with respect to a network structure changing, is thus improved. We show the effectiveness of our algorithm on a grid network, where the vector of the travel times through the links of the network is assumed to have a multivariate gamma probability distribution, taking into account the correlations of the travel times. We present preliminary results of the application of the proposed guidance strategy, with comparisons to the existing algorithm. We show, in particular, the variance of the calculated optimal paths and strategies, with respect to the users' sensitivity towards the existence and the performance of alternative detours of the selected paths. The model is then able to calculate robust guidance strategies towards path failures.