In many future joint-action scenarios, humans and robots will have to interact physically in order to successfully cooperate. Ideally, seamless human-robot interaction should not require training for the human, but should be intuitively simple. Nonetheless, seamless interaction and cooperation involve some degree of learning and adaptation. Here, we report on a simple case of physical human-robot interaction, a hand-over task. Even such a basic task as manually handing over an object from one agent to another requires that both partners agree upon certain basic prerequisites and boundary conditions. While some of them are negotiated explicitly, e.g. by verbal communication, others are determined indirectly and adaptively in the course of the cooperation. In the present study, we compared human-human hand-over interaction with the same task done by a robot and a human. To evaluate the importance of biological motion, the robot human interaction was tested with two different velocity profiles: a conventional trapezoidal velocity profile in joint coordinates and a minimum-jerk profile of the end-effector. Our results show a significantly shorter reaction time for minimum jerk profiles, which decreased over the first three hand-overs. The results of our comparison provide the background for implementing effective joint-action strategies in humanoid robot systems.

We present the exploring/exploiting tree (EET) algorithm for motion planning. The EET planner deliberately trades probabilistic completeness for computational efficiency. This tradeoff enables the EET planner to outperform state-of-the-art sampling-based planners by up to three orders of magnitude. We show that these considerable speedups apply for a variety of challenging real-world motion planning problems. The performance improvements are achieved by leveraging work space information to continuously adjust the sampling behavior of the planner. When the available information captures the planning problem's inherent structure, the planner's sampler becomes increasingly exploitative. When the available information is less accurate, the planner automatically compensates by increasing local configuration space exploration. We show that active balancing of exploration and exploitation based on workspace information can be a key ingredient to enabling highly efficient motion planning in practical scenarios.

This paper presents a concept of a smart working environment designed to allow true joint-actions of humans and industrial robots. The proposed system perceives its environment with multiple sensor modalities and acts in it with an industrial robot manipulator to assemble capital goods together with a human worker. In combination with the reactive behavior of the robot, safe collaboration between the human and the robot is possible. Furthermore, the system anticipates human behavior, based on knowledge databases and decision processes, ensuring an effective collaboration between the human and robot. As a proof of concept, we introduce a use case where an arm is assembled and mounted on a robotpsilas body.

Human activity recognition is crucial for intuitive cooperation between humans and robots. We present an approach for activity recognition for applications in the context of human-robot interaction in industrial settings. The approach is based on spatial and temporal features derived from skeletal data of human workers performing assembly tasks. These features were used to train a machine learning framework, which classifies discrete time frames with Random Forests and subsequently models temporal dependencies between the resulting states with a Hidden Markov Model. We considered the following three groups of activities: Movement, Gestures, and Object handling. A dataset has been collected which is comprised of 24 recordings of several human workers performing such activities in a human-robot interaction environment, as typically seen at small and medium-sized enterprises. The evaluation shows that the approach achieves a recognition accuracy of up to 88% for some activities and an average accuracy of 73%.