Highly heterogeneous problems arise in many scientific and engineering applications, such as in the simulation of composite materials, groundwater flow, or even topology optimization. Unfortunately, the resulting jumps in the distribution of material coefficients often deteriorate the convergence of iterative solvers, such as multi-level domain decomposition methods. One remedy is to enhance the coarse space of the solvers via additional basis functions. A popular and robust approach is the use of adaptive/spectral coarse spaces [1,2], in which the coarse basis functions are computed via local generalized eigenvalue problems (GEVPs). While the locality of the GEVPs limits both the required communication to neighbouring subdomains and the size of the GEVPs, the computational overhead posed by setup and solution of the GEVPs can still become a performance bottleneck. We introduce a new robust coarse space for the dual-primal finite element tearing and interconnecting (FETI-DP) method. In contrast to classical heuristic approaches that build the basis function without solving GEVPs, our approach adopts the reduced basis approach to significantly accelerate solution of GEVP. While the robustness of this approach has not been established theoretically yet, the numerical results demonstrate robustness comparable to theoretically-founded adaptive/spectral coarse spaces while significantly improving the computational efficiency.