A $k$-submodular function is a generalization of the submodular set function. Many practical applications can be modeled as maximizing a $k$-submodular function, such as multi-cooperative games, sensor placement with $k$ type sensors, influence maximization with $k$ topics, and feature selection with $k$ partitions. In this paper, we provide a unified continuous greedy algorithm for $k$-submodular maximization problem under a down-monotone constraint. Our technique involves relaxing the discrete variables in a continuous space by using the multilinear extension of $k$-submodular function to find a fractional solution, and then rounding it to obtain the feasible solution. Our proposed algorithm runs in polynomial time and can be applied to both the non-monotone and monotone cases. When the objective function is non-monotone, our algorithm achieves an approximation ratio of $(1/e-o(1))$; for a monotone $k$-submodular objective function, it achieves an approximation ratio of $(1-1/e-o(1))$.