An intelligent cane robot is designed for aiding the elderly and handicapped people's walking. The robot consists of a stick, a group of sensors, and an omnidirectional basis driven by three Swedish wheels. Recognizing the user's walking intention plays an important role in the motion control of our cane robot. To quantitatively describe the user's walking intention, a concept called “intentional direction (ITD)” is proposed. Both the state model and the observation model of ITD are obtained by enumerating the possible walking modes and analyzing the relationship between the human-robot interaction force and the walking intention. From these two models, the user's walking intention can be online inferred using the Kalman filtering technique. Based on the estimated intention, a new admittance motion control scheme is proposed for the cane robot. Walking experiments aided by the cane robot on a flat ground and slope are carried out to validate the proposed control approach. The experimental results show that the user feels more natural and comfortable when our intention-based admittance control is applied.