Title: Human-inspired release controller for natural robot to human handover tasks Abstract : One of the challenges faced by the robotics research community is to develop a robot assistant able to collaborate with humans in domestic and working scenarios. The most basic collaborative motor task between humans is probably the object handover, in which an object is passed between two agents: the passer and the receiver. Object handover is fundamental for a wide range of functional or social activities in which humans help each other sharing the same goal and a common plan of execution. The way in which an object is released by the hand of a robotic passer plays a fundamental role in the perceived fluency of the handover action by the human receiver. Behavioral studies in humans have highlighted how during object handover, the passer and the receiver regulate their grip forces in complementary fashion in order to accomplish a successful transfer. The initial contact with the receiver is largely controlled in a feed-forward manner by the passer who produces a stereotypic movement defined using visual cues describing the partner’s movements, as well as on the experience/practice of the passer. After contact the passer grip force decreases following a characteristic profile within a certain time window. Such behavior anticipates the likely action of the receiver in taking the object based on previous experience of the handover action. With the aim to increase the comfort and the perceived fluency of the human robot interaction during object handover, we investigate a two parameter feed-forward controller with human-like dynamics for robotic hands able to release objects and tools. Once triggered by contact between the receiver and the object held by the robot hand, the grip force of the robotic passer is gradually reduced to zero, following a stereotypical and time-based trajectory. We compared the controller, tuned with a range of parameters, with the human-human handover and with a state-of-the-art feedback controller proposed in the literature. In addition we evaluated the quality of the interactions as rated on a questionnaire completed by eight participants. Results show that the control strategy proposed in this paper is consistently superior to the state-of-the-art mechanism and that feed-forward control strategies with a lower threshold on force to trigger the release are generally preferred. We are currently testing a new controller able to combine the benefits of the feedforward strategy (i.e. perception of the fluency of the handover) and of the feedback strategy (i.e. adaptation of the handover dynamics to the receiver uncertainties).