Hebbian Plasticity in CPG Controllers Facilitates Self-Synchronization for Human-Robot Handshaking

Abstract

It is well-known that human social interactions generate synchrony phenomena which are often unconscious. If the interaction between individuals is based on rhythmic movements, synchronized and coordinated movements will emerge from the social synchrony. This paper proposes a plausible model of plastic neural controllers that allows the emergence of synchronized movements in physical and rhythmical interactions. The controller is designed with central pattern generators (CPG) based on rhythmic Rowat-Selverston neurons endowed with neuronal and synaptic Hebbian plasticity. To demonstrate the interest of the proposed model, the case of handshaking is considered because it is a very common, both physically and socially, but also, a very complex act in the point of view of robotics, neuroscience and psychology. Plastic CPGs controllers are implemented in the joints of a simulated robotic arm that has to learn the frequency and amplitude of an external force applied to its effector, thus reproducing the act of handshaking with a human. Results show that the neural and synaptic Hebbian plasticity are working together leading to a natural and autonomous synchronization between the arm and the external force even if the frequency is changing during the movement. Moreover, a power consumption analysis shows that, by offering emergence of synchronized and coordinated movements, the plasticity mechanisms lead to a significant decrease in the energy spend by the robot actuators thus generating a more adaptive and natural human/robot handshake.

Motor Coordination Learning for Rhythmic Movements

“…This is similar to what we observed in on-going research of human interactions where individuals perform the same movement at the same frequency but each maintaining their own characteristics. Moreover, in our previous case study of handshaking [24], the CPG was entrained due to the physical locking. In this case, there is no physical contact so plasticity mechanisms truly are responsible for the motor coordination and adaptation.…”

“…Note also that in [24], we used force feedback and the CPG output as a velocity command. Here, as there is no physical interaction and the Pepper robot has neither velocity control nor torque sensors, a similar setup could not possibly be done.…”

“…The output of the CPG is thus considered as an angular position offset and the position control mode to command the joints of Pepper is employed. For a better understanding of this subsection, please refer to [24] where the CPG model is extensively detailed.…”

“…The learning rules proposed in [24] can be applied and frequency learning, inspired by [25], is defined as:…”

CPG-based Controllers can Generate Both Discrete and Rhythmic Movements

“…σ s is a gain. To be included in the proposed CPG architecture [22], equations of the rhythmic cells can be rewritten as (please refer to [29] for all the details):…”