Final Report Summary - SHRINE (Seamless Human Robot Interaction in Dynamic Environments)
Robots seamlessly acting in joint workspaces with humans constitutes the leading vision of the SHRINE project. Hence, the largest share of effort within the project concentrated on the intersection of socio-psychological and engineering sciences.
Based on observations in human-human interactions, socio-psychological human-robot interaction mechanisms conveyed by speech and facial expressions were developed. Evaluations with an expressive robot head showed increased user acceptance and the willingness to cooperate and interact with robots. Most important in this context was the introduction of a behavior control model that enables robots to successfully trigger empathy and prosocial behavior in human users towards a robot. Further improvement of user acceptance was achieved in the field of navigation in human populated indoor and outdoor environments with a special focus on approaching humans. A motion planning algorithm respecting social guidelines was proposed, based on optimal control principles.
In order to regard and accept robots as equal entities their manipulation skills should be able to compete with a wide range of human manipulation skills. In this context, the project focused on identifying and solving fundamental skills with generic application potential. The use of optimal control methods and contributions within hybrid dynamical system theory enabled fast manipulation of objects with complex geometry while being restricted to simple robots and end-effectors. At the same time, the non-complex hardware setup allows exact and unique analytic solutions to the fundamental manipulation skills such that explicit bounds for task success could be derived. Beyond that, the SHRINE framework brings together these advances in dynamic manipulation with the newly introduced research field of human-robot cooperative dynamic (object) manipulation. Experiments confirmed, for example, that with an automatically adapting controller pure haptic feedback is sufficient and only little model knowledge is required to cooperatively inject energy into unknown flexible objects in order to potentially place it outside the agents’ workspace.
Another essential class of skills humans make use of is the nonverbal exchange of information. In the research introduced above this aspect was also covered in the human-robot cooperation. By using observers based on the fundamental dynamics, the use of leader and follower controllers were enabled. As a result, the robot is able to infer the desired task goal of the human partner. Planning motions in a shared dynamic environment is another example highly depending on nonverbal information exchange. By analyzing human walking behavior, e.g. with a game theoretic approach, the human-likeness of robot motions was improved, ultimately leading to better predictability of the robotic entity and thus improving its acceptance. Furthermore, the partially structured nature of such shared environments was formalized in a framework that advanced the state of art from the prediction of individual agents to a deterministic and feasible approach that captures the entire current environment situation.
In total, the above contributions have brought the vision of robots seamlessly acting in joint workspaces with humans much closer. Hereby, the interdisciplinary approach in this project has shown to be a key factor in order to alleviate barriers lying between today’s robots and their future of being equal entities.
Based on observations in human-human interactions, socio-psychological human-robot interaction mechanisms conveyed by speech and facial expressions were developed. Evaluations with an expressive robot head showed increased user acceptance and the willingness to cooperate and interact with robots. Most important in this context was the introduction of a behavior control model that enables robots to successfully trigger empathy and prosocial behavior in human users towards a robot. Further improvement of user acceptance was achieved in the field of navigation in human populated indoor and outdoor environments with a special focus on approaching humans. A motion planning algorithm respecting social guidelines was proposed, based on optimal control principles.
In order to regard and accept robots as equal entities their manipulation skills should be able to compete with a wide range of human manipulation skills. In this context, the project focused on identifying and solving fundamental skills with generic application potential. The use of optimal control methods and contributions within hybrid dynamical system theory enabled fast manipulation of objects with complex geometry while being restricted to simple robots and end-effectors. At the same time, the non-complex hardware setup allows exact and unique analytic solutions to the fundamental manipulation skills such that explicit bounds for task success could be derived. Beyond that, the SHRINE framework brings together these advances in dynamic manipulation with the newly introduced research field of human-robot cooperative dynamic (object) manipulation. Experiments confirmed, for example, that with an automatically adapting controller pure haptic feedback is sufficient and only little model knowledge is required to cooperatively inject energy into unknown flexible objects in order to potentially place it outside the agents’ workspace.
Another essential class of skills humans make use of is the nonverbal exchange of information. In the research introduced above this aspect was also covered in the human-robot cooperation. By using observers based on the fundamental dynamics, the use of leader and follower controllers were enabled. As a result, the robot is able to infer the desired task goal of the human partner. Planning motions in a shared dynamic environment is another example highly depending on nonverbal information exchange. By analyzing human walking behavior, e.g. with a game theoretic approach, the human-likeness of robot motions was improved, ultimately leading to better predictability of the robotic entity and thus improving its acceptance. Furthermore, the partially structured nature of such shared environments was formalized in a framework that advanced the state of art from the prediction of individual agents to a deterministic and feasible approach that captures the entire current environment situation.
In total, the above contributions have brought the vision of robots seamlessly acting in joint workspaces with humans much closer. Hereby, the interdisciplinary approach in this project has shown to be a key factor in order to alleviate barriers lying between today’s robots and their future of being equal entities.