While walking around a city, the temporary obstacles present on the sidewalk barely register in most people’s minds. The reality for people with disabilities is quite different, whether it’s a scooter left in the way, crowds that refuse to budge, or construction work loud enough to trigger somatic effects. While detecting permanent obstacles (e.g. wheelchair-inaccessible areas) is a relatively easy thing, detecting and addressing temporary obstacles is very difficult. The objective of this paper is to propose some first elements to build a framework aimed at detecting temporary obstacles for diversely disabled users. We point out several scientific and technical issues that pave the way to reach this goal and highlight the limits of existing approaches. We insist on three signif- icant problems to overcome: incomplete models of the environment, lim- ited availability of good-quality data, and absence of tailored algorithms. Taking inspiration from percolation theory, we propose some leads to solve the first two problems mentioned.