Fast Handovers with a Robot Character: Small Sensorimotor Delays Improve Perceived Qualities Matthew K.X.J. Pan∗;1, Espen Knoop∗;2, Moritz Bacher¨ 2 and Gunter¨ Niemeyer1 Abstract— We present a system for fast and robust handovers with a robot character, together with a user study investigating the effect of robot speed and reaction time on perceived interaction quality. The system can match and exceed human speeds and confirms that users prefer human-level timing. The system has the appearance of a robot character, with a bear-like head and a soft anthropomorphic hand and uses Bezier´ curves to achieve smooth minimum-jerk motions. Fast timing is enabled by low latency motion capture and real- time trajectory generation: the robot initially moves towards an expected handover location and the trajectory is updated on-the-fly to converge smoothly to the actual handover loca- tion. A hybrid automaton provides robustness to failure and unexpected human actions. In a 3x3 user study, we vary the speed of the robot and add variable sensorimotor delays. We evaluate the social perception of the robot using the Robot Social Attribute Scale (RoSAS). Inclusion of a small delay, mimicking the delay of the human sensorimotor system, leads to an improvement in perceived qualities over both no delay and long delay conditions. Specifically, with no delay the robot is perceived as more discomforting, and with a long delay it is perceived as less warm. Fig. 1. The robotic character can perform both human-to-robot and robot- to-human handovers. I. INTRODUCTION Robots are starting to interact directly with humans and are gradually becoming more involved in our daily social interactions were significantly slower than typical human- interactions—as helpers, companions, and care-givers. This human handovers. means that robots are not only required to be safe and To further explore this area, this work presents a system functional, but should also act consistent with normal and that is capable of executing fast and robust bidirectional expected human behaviors. (human-to-robot and robot-to-human) handovers. We believe Handing over an object requires little conscious thought matching and exceeding human timing more readily allows for a human, yet is filled with expectations and is seen users to anthropomorphize and perceive the robot as part of against a lifetime of experiences. For robots, it presents a a normal social interaction. relevant example of a direct interaction with a human, and As part of an effort to encourage such perception, the robot thus, handover interactions between humans and robots have has the appearance of the torso of a bear-like character and been a topic of much study. We are particularly interested in features a head and an anthropomorphic hand as seen in endowing robots with handover behaviors that users perceive Fig. 1. Rounding out this compelling robotic character, the favorably and as natural, competent, and efficient. system uses adjusted minimum-jerk movements and is robust In previous work [1], we studied handover interactions towards unexpected or uncooperative human behaviors. with a non-anthropomorphic robot and identified timing as We use this robot character handover system to conduct the factor with the greatest effect on perceived qualities: a 3x3 user study where we vary the speed of the robot faster behaviors were preferred over slower ones. We hypoth- motions and the system reaction time. To vary the reaction esized that participants preferred interactions that were more time, we include a variable sensorimotor delay. From our efficient, i.e., required less time. However, all the studied previous work [2], we know that closed-loop control with a small sensorimotor delay was preferred in human-robot This work has been supported by Disney Research, and by the European handshaking. We thus hypothesize that the inclusion of a Commission Horizon 2020 Framework Programme (Soma project, grant H2020-ICT-645599). delay that mimics the latency of the human sensorimotor ∗The first two authors contributed equally. system will create a more compelling behavior. We consider 1 MP and GN are with Disney Research, Glendale, CA. three levels of sensorimotor reaction time: no delay (faster- [email protected], [email protected] 2EK and MB are with Disney Research, Zurich, Switzerland. than-human reaction), short delay (similar to human reaction [email protected], [email protected] time) and long delay (slower than human reaction time). For the speed, we consider three levels—slow, moderate, and III. SYSTEM FOR FAST AND ROBUST HANDOVER fast—of which the moderate condition is similar to the speed INTERACTIONS of human arm motions. A. Handover Task Our study results show that the inclusion of the short This paper considers a bidirectional handover interaction. sensorimotor delay improves the perceived qualities of the The human initiates by presenting the object. The robot robot. With no delay added, the system is perceived as more reaches, grasps, and moves the object from the handover discomforting—independently of the arm speed conditions location to its resting position. It then returns the object to the tested. With a long delay, the robot is perceived as less warm. same handover location. The handover sequence is depicted In the following, we present the system as well as the user in Fig. 2. study followed by remarks discussing the work. The handovers are performed with a toroidal object of 30 cm diameter, depicted in Fig. 3. The toroidal shape of the object can readily be grasped by the robot from a range of approach angles. It also distances the human II. RELATED WORK from the robot hand, eliminating potential interference and improving safety. An OptiTrack motion capture system uses a constellation of retroreflective markers on the toroid to Handover interactions have received significant attention track its position and orientation. Users do not wear markers in the robotics literature. This includes both the characteriza- or other instrumentation. During the user study, the object tion of human-human interactions to gain detailed models of is initially placed in a cradle within reach of the human different aspects of handover interactions [3, 4, 5, 6], as well participant. The handover interaction starts when the object as their implementation on robotic systems. The majority of is removed from the cradle. these robotic implementations have focused on the robot- B. Robot Character to-human handover direction [7, 8, 9, 5, 10], with some work considering human-to-robot handovers or bidirectional To aid the perception of the robot as a social entity, interactions [4, 11, 12, 13, 14]. Prior work has considered the we create the appearance of an anthropomorphic bear-like robustness of handover systems to failure. In [15], a robot character with torso, arm, and head. The system uses a is fitted with an object acceleration sensing setup such that KUKA LBR iiwa 7 R800 robot mounted horizontally to form the robot can re-grasp a falling object during handover. In the shoulders and right arm of the bear character as shown in another stream of work, the role of gaze in handover interac- Fig. 3. In this way, joint 6 of the robot becomes the elbow tions has been studied, showing that handover performance of the character, and joint 4 becomes the character’s right is improved if the robot uses its gaze as a cue to the user shoulder. A cartoon bear head is attached to the second link [16, 17]. and the “torso” is dressed in a shirt to reinforce the illusion of a character handover. While robotic handover systems frequently implement We use a Pisa/IIT SoftHand [25], which is a soft and findings from human-human interactions, these systems are underactuated hand with a single actuated degree of freedom. in general substantially slower than the human-human coun- The hand has a five-fingered anthropomorphic design, which terparts they emulate [18, 12]. Yet, timing in handover inter- supports the character’s appearance. Moreover, its softness actions has been posed as an important aspect by multiple allows the hand to robustly grasp in the presence of small works. Koene et al. [19] study the relative importance of locational variations of the object. spatial and temporal precision for handover interactions, and To enhance the character behavior, joint 1 of the robot tilts show that the temporal aspects are of greater importance. the head to appear to look at the object. Joint 2 allows the Hoffman [20] discusses the importance of using fluency as character to lean forward with its right shoulder to reach an evaluation metric for human-robot interactions. In our towards a more distant handover location, i.e. beyond a previous work, we found timings to play a significant role specified radius of its right shoulder, or leans back if the in the social perception of handover interactions [1]. Also location is too close. relevant, Admoni et al. [21] study the effect of introducing An analytic inverse kinematics solver computes the re- delays to emphasize robot non-verbal communication such maining five joints to grasp the toroid. With two axes of as gaze. symmetry around the major and minor radii, the toroid The social perception of robots in handovers and in other strictly requires only four degrees of freedom to achieve a contexts has received some focused study in recent years: grasp. We thus use the final degree of freedom to keep the Aleotti et al. [22] argue for the importance of robots behaving elbow height as low as possible. in a social manner when interacting with humans. The Robot Social Attribute Scale (RoSAS) [23] has been presented C. Online Trajectories for Fast and Smooth Motions recently as a psychometrically-validated scale for measuring Human receivers often begin reaching for the object before the social perception of robots.