Interacting humans and robots can improve sensory prediction by adapting their viscoelasticity

TL;DR

A computational model of the mechanical and sensory interactions between agents that can tune their viscoelasticity while considering their sensory and motor noise is developed and the resulting stochastic-optimal-information-and-effort (SOIE) controller predicts how the exchange of haptic information and the performance can be improved by adjusting viscoelasticity.

Abstract

To manipulate objects or dance together, humans and robots exchange energy and haptic information. While the exchange of energy in human-robot interaction has been extensively investigated, the underlying exchange of haptic information is not well understood. Here, we develop a computational model of the mechanical and sensory interactions between agents that can tune their viscoelasticity while considering their sensory and motor noise. The resulting stochastic-optimal-information-and-effort (SOIE) controller predicts how the exchange of haptic information and the performance can be improved by adjusting viscoelasticity. This controller was first implemented on a robot-robot experiment with a tracking task which showed its superior performance when compared to either stiff or compliant control. Importantly, the optimal controller also predicts how connected humans alter their muscle activation to improve haptic communication, with differentiated viscoelasticity adjustment to their own sensing noise and haptic perturbations. A human-robot experiment then illustrated the applicability of this optimal control strategy for robots, yielding improved tracking performance and effective haptic communication as the robot adjusted its viscoelasticity according to its own and the user's noise characteristics. The proposed SOIE controller may thus be used to improve haptic communication and collaboration of humans and robots.

Figures (6)

• Figure 1: Human and robot agents collaborating on a tracking task while exchanging energy and haptic information through an elastic band. (A) Each agent perceives the target movement and their own position, as well as the motion of their partner through the interaction force. The visual and haptic information are perturbed by their respective sensory noise. (B) Each of these soft agents can adapt their viscoelasticity $\{L,\tilde{L}\}$ for following their own planned trajectory to track the target, while considering the partner's movement toward the same target as perceived from the interaction. (C) Protocol of human-robot experiment. Participants started with an initial 10 solo trials to familiarise themselves with the task and interface dynamics, followed by 4 blocks of 10 interaction trials with one controller, and another 4 blocks of 10 trials with the other controller.
• Figure 2: Impedance adaptation in robot-robot interaction. (A) How optimal impedance varies with one's own and with partner noise. (B) The effort difference between the SOIE computed impedance and the use of low/high impedance. (C) The perceived target decoded from the interaction torque. With optimal impedance this becomes more similar to the real target trajectory. (D) The tracking error difference. (E) The prediction of the target trajectory by integrating one's own sensing and the interaction with a partner: optimal impedance yields prediction that is superior to both partners with biased and unbiased observations.
• Figure 3: Prediction of performance and cocontraction (shown with 'x's) in human-human interaction compared to experimental data (shown as intervals defined by the mean value $\pm$ standard deviation) , where SN stands for sharp (self)-noisy (partner), etc. (A) Tracking error with respect to the different conditions. (B) Predicted muscle cocontraction with experimental data. Statistical analysis comparing the differences in experimental data is shown where ** $p<0.01$, *** $p<0.001$.
• Figure 4: Tracking results for the human-robot collaboration experiment. (A) The 12 participants’ tracking errors and effort in the last four trials of all the conditions. The differences in the average normalised costs (B) and tracking error (C) of the robot agent for the SOIE controller and high impedance control (HIC) when the robot had noisy sensing conditions. (D) The predicted target movement. This is more accurate with the SOIE controller compared to the HIC when integrating the interaction torque from human partners. The width of each strip represents the standard deviation of the target prediction among the 12 participants.
• Figure S1: Tracking errors and muscle cocontraction of SOIE (left) and high impedance controller (right) in the human-robot experiment. (A) Evolution of the tracking error over the 12 participants charted as a function of the trials, where the error bars represent one standard error on each side. When the robot agent has sharp sensory information, the tracking errors are similarly small for both controllers with little influence from human visual noise. However, when the robot agent has noisy sensory information, human visual noise has a larger impact on the tracking performance. Optimal impedance slightly improves the tracking in the NN condition as the error consistently decreases along trials while high impedance saturates has an increasing trend in after the first three trials. (B) Evolution of the normalized cocontraction as a function of trials, where the error bars represent one standard error. Human effort is reduced when the robot has noisy sensory information (especially in the NS condition), providing effective assistance to human users.