In order for a human to interact intuitively with a robot, the controller must be designed carefully. Indeed, traditional position control schemes are not suited for physical human-robot interaction.

    An appropriate scheme is admittance control. It consists in measuring the forces applied on the robot through a force/torque sensor and computing the robot reaction corresponding to a dynamic defined by virtual mass, damping and stiffness. For ergonomic purposes, the user feels a small virtual mass even if the robot which is collaborating with him is carrying a heavy load. Figure 1 presents a human interacting with a serial robot through a force/torque sensor.

    Fig. 1: Example of physical human-robot interaction using a force/torque sensor.

    It is also possible to interact with a robot by applying forces directly on one of its links if it is covered with a tactile sensor, as can be seen in Figure 2. This type of physical collaboration is more intuitive because it is similar to an interaction between two humans. However, current robotic skins do not measure the shearing forces applied on them. For this reason, it is easier to implement a human-robot interaction control scheme using a force/torque sensor. However, combining both measurement systems allow the simultaneous control of the end-effector's motion and the additional degrees of freedom of a redundant robot.

    Fig. 2: Example of physical human-robot interaction using a robotic skin.

    In order to improve the interaction, it is possible to modify the virtual dynamics online. An interesting approach consists in the increase of the virtual damping factor when the human wants to decelerate the robot. Indeed, it is frequently required to stop a load rapidly after a gentle acceleration. To detect the human intent to decelerate, the directions of the velocity and the applied force are compared. A force opposed to the displacement is considered as a will to stop the robot. Figure 3 presents a variable admittance control scheme.

    Fig. 3: Control scheme using a variable admittance model.

    It can be observed on the control scheme that the human is in the control loop. Because each human is different, this explains the numerous challenges to improve the control of interactive robots! These include the instability of the control system which happens when the human stiffens his arms. To avoid these dangerous situations, an instability observer has been developed, based on the Lyapunov theory of stability. In practice, the human's stiffness is estimated using force and displacement measurements. If it exceeds a certain threshold, the controller responds by increasing the virtual damping. Even if it is possible to always use a large virtual damping, it is preferable to modulate it in order for the collaboration to be ergonomic, intuitive and stable, for any user behaviour. The second video at the end of this page presents the experimental behaviour obtained by using an instability observer.