Human-Following Experiment

The results of the experiment are shown in Fig.11(a). The control parameters were tuned empirically to $M=30.0[kg]$, $I=1.5[kg{\cdot}m^2]$, $k_1=3.0[N/m]$, $k_2=0.001[N{\cdot}rad]$, $k_3=30[N{\cdot}s/m]$, $k_4=20.0[ts/rad]$, and $l_0$ is set to 60$[cm]$. A human walked along a hexagonal path. The mobile robot followed him along the course plotted by four kinds of signs in Fig.11(a). Fig.11(b) shows the velocity control inputs for the mobile robot from the DINDs in this experiment.

In spite of the unstable human position estimation, the mobile robot was controlled smoothly and there were no abrupt changes in the control inputs. The low frequency fluctuations in the human trajectory were cut off by the virtual spring and the robot was controlled smoothly. It is found from the results that the proposed algorithm allows the robot to follow the target without excessive motions.

In Fig.11(a), the dashed line represents the position measured by the encoders in the robot. The dashed line deviates significantly from the trajectory of the robot measured by the DINDs. This is mainly caused by slips of the wheels and the wrong internal parameters of the robot. It is also an advantage for the human-following robot realized in ISpace to be robust against various errors that disturb the robot's navigation.

Although this human path is simple from a global point of view, complex motions, such as variations in the velocity and direction of humans are included in the human's actual movement. The validity of the proposed control method was verified from this experimental result.

[Trajectories of the robot and human] (a)

[Input velocity] (b)

図 11: Experimental results