In physical human-robot interaction (pHRI) tasks, the control of interaction forces by each partner is critical. While robots can be programmed to independently control force and motion [3], two recent studies demonstrated that humans cannot decouple their control of force from motion when collaboratively tracing a path with a robot partner. Maurice and colleagues showed that humans applied significant interaction forces against the robot’s predefined elliptic path, despite being instructed to minimize any interaction forces [2]. A follow-up study demonstrated that humans were unable to maintain a constant instructed force during the robot’s motion, but exhibited periodic force modulations around the elliptic path [5]. As these findings highlighted important limitations for humans collaborating with robots, a third study on the same ellipse tracing task showed that augmented feedback enabled humans to better regulate their interaction forces. However the forces were still significant even after 3 days of practice [1]. The present study aims to identify the specific features of the robot motion that lead to the lingering undesired interaction forces in the physical human- robot interaction task. This study used the same ellipse tracing task in which the robot’s trajectory around the elliptic path obeyed or violated the so-called 2/3 power law in human trajectory control[4]: humans tend to slow down in highly curved path segments and speed up in straighter segments. Based on this 2/3 power law, a robot was programmed to move with three distinct velocity profiles around an ellipse in the horizontal plane. One profile followed the 2/3 power law, consistent with human preferences. Two additional profiles challenged participants to deviate from their preferred velocity: one condition exaggerated the human-like velocity modulation, another condition kept tangential velocity constant. We tested three specific candidates that may elicit undesired forces: curvature (Hypothesis 1, H1), tangential velocity (H2), or angular velocity (H3) of the movement. Our findings will help understand the limitations of human motor control to inform robot controllers that can better accommodate the human partner.