Evaluation by Expert Dancers of a Robot That Performs Partnered Stepping via Haptic Interaction

Abstract

Our long-term goal is to enable a robot to engage in partner dance for use in rehabilitation therapy, assessment, diagnosis, and scientific investigations of two-person whole-body motor coordination. Partner dance has been shown to improve balance and gait in people with Parkinson's disease and in older adults, which motivates our work. During partner dance, dance couples rely heavily on haptic interaction to convey motor intent such as speed and direction. In this paper, we investigate the potential for a wheeled mobile robot with a human-like upper-body to perform partnered stepping with people based on the forces applied to its end effectors. Blindfolded expert dancers (N=10) performed a forward/backward walking step to a recorded drum beat while holding the robot's end effectors. We varied the admittance gain of the robot's mobile base controller and the stiffness of the robot's arms. The robot followed the participants with low lag (M=224, SD=194 ms) across all trials. High admittance gain and high arm stiffness conditions resulted in significantly improved performance with respect to subjective and objective measures. Biomechanical measures such as the human hand to human sternum distance, center-of-mass of leader to center-of-mass of follower (CoM-CoM) distance, and interaction forces correlated with the expert dancers' subjective ratings of their interactions with the robot, which were internally consistent (Cronbach's α=0.92). In response to a final questionnaire, 1/10 expert dancers strongly agreed, 5/10 agreed, and 1/10 disagreed with the statement "The robot was a good follower." 2/10 strongly agreed, 3/10 agreed, and 2/10 disagreed with the statement "The robot was fun to dance with." The remaining participants were neutral with respect to these two questions.

Fig 1. Experimental setup.

An expert dancer leads the robot Cody during partnered stepping.

Fig 2. Biomechanics of human-robot partnered stepping.

Example data from two cycles of one trial from one participant. Gray and white bars indicate intervals of time when right and left feet were on the ground, respectively. The experimental treatment for this trial was low gain, low stiffness.

Fig 3. Correlation between subjective and objective measures.

Biomechanical measures are listed on the rows and subjective measures are listed on the columns in descending order of number of significant correlations. Black and white boxes denote significant negative and positive Pearson’s correlations coefficients, respectively. Gray denotes non-significant correlations. For example, with increasing lag, participants report that the robot follower moves and understands the leader’s motor intention less well. Also, as the interaction force increases, the participants report that the robot follower becomes less easy to communicate or move with. Similarly, as the variability in hand-sternum / CoM-CoM distance increases, the follower understands or moves according to the leader’s motor intention less accurately. Interestingly, cadence variability, RMS, and MSE values have very little correlation with any of the subjective responses. Such useful insights are possible through this correlation matrix between the biomechanical measures and subjective responses.

Fig 4. Final questionnaire responses regarding overall experience.

Response level of 1 = “Strongly Disagree,” 3 = “Neutral,” 5 = “Strongly Agree.” p-values show results from one-sample t-tests comparing with a response level of 3.

Fig 5. High admittance gain results in higher subjective dance performance.

(A) Motor intent, (B) Motor performance, (C) Motor skill. Bars show mean and standard error. Response level of 1 = “Strongly Disagree,” 3 = “Neutral,” 5 = “Strongly Agree.”

Fig 6. Humans adapt force input to maintain constant velocity.

(A) Humans exert 0.53x and 0.56x force at the high gain setting compared to the low gain setting when walking forward and backward, respectively. (B) Humans maintain similar velocities across all conditions. Bars show mean and standard error.

Fig 7. Biomechanical measures according to gain and stiffness.

(A) Lag time of robot behind human, (B) CoM-CoM distance, (C) CoM-CoM distance standard deviation, (D) human left hand to sternum distance, (E) human left hand to sternum distance standard deviation. Bars show mean and standard error.

Fig 8. Acrylic rig used in the high stiffness condition.

Black cloth sleeves were draped over the robot’s upper arms to conceal the presence or absence of the acrylic rig during the high stiffness and low stiffness conditions, respectively.

Fig 9. Model of the robotic system.

The damper with damping coefficient b corresponds with the mobile base and the spring with spring constant k corresponds with the robot’s arm. F and $\stackrel{.}{x}$ are the force and velocity at the robot’s end effector.

Fig 10. Bode plot for Low Gain, Low Stiffness.

Input and output are force and velocity at the end effector, respectively. Empirical curve shows measured response of the robot. Theoretical curve shows the response of the ideal spring-damper model.