On the feasibility of learning to predict minimum toe clearance under different walking speeds