Adaptive human force scaling for physical human-robot interaction via admittance control

Collaborative robotic systems are designed to be lightweight, portable and are equipped with inherent safety features. These qualities allow such robots to be situated in close proximity to humans, and even enable daily activities that require direct physical interaction with humans, such as object co-manipulation. Co-manipulation is prevalent in many industries, such as production, manufacturing, and construction, which still rely on manual labour of skilled manpower. Manipulating bulky and/or heavy objects are ergonomically hard, hence are typically handled by two humans in those industries, even though they are non-value-added tasks. Therefore, a collaborative robot can be introduced to assist the human partner in such manipulation tasks in order to decrease the human effort and save labor time. Nevertheless, if the robot is programmed to passively follow human movements, it would be an extra burden to the human partner, particularly when human movement intentions change during the task. On the other hand, if the robot executes the task based on pre-programmed plans without paying attention to the human partner's intentions, conflicts will arise, and if these conflicts are not resolved, the task will be exhausting for the human. Hence, for the robot to perform a collaborative manipulation task naturally and effectively, it has to be aware of human intentions and act accordingly. In the first part of this dissertation, we design an admittance controller for a robot to adaptively change its contribution to a collaborative manipulation task executed with a human partner to improve the task performance. This has been achieved by adaptive scaling of human force based on her/his movement intention while paying attention to the requirements of different task phases. In our approach, movement intentions of human are estimated from measured human force and velocity of manipulated object and converted to a quantitative value using a fuzzy logic scheme. This value is then utilized as a variable gain in an admittance controller to adaptively adjust the contribution of robot to the task without changing the admittance time constant. We demonstrate the benefits of the proposed approach by a pHRI experiment utilizing Fitts' reaching movement task. The results of the experiment show that there is a) an optimum admittance time constant maximizing the human force amplification and b) a desirable admittance gain profile which leads to a more effective co-manipulation in terms of overall task performance. In the second part of this dissertation, we propose a machine learning (ML) approach to resolve conflicts that may occur between human and a proactive robot during the execution of a collaborative task. We suggest focusing on the dyadic interaction patterns to detect and resolve such conflicts. We further argue that the features derived from haptic (force/torque) data only as inputs to a ML algorithm is sufficient to classify if human and robot manipulate the object harmoniously or they face a conflict. In our approach, the robot contributes to the task proactively when there is harmony or follows human behavior passively when there is a conflict. We implement an admittance controller to regulate the physical interaction between human and robot during the task (hence, the robot follows the human passively when there is a conflict) and utilize an artificial potential field approach to make the robot proactive when the partners work in harmony. We designed experimental scenarios involving harmonious and conflicting interactions between human and robot during a collaborative manipulation of an object to train and test our ML model. The results of the study show that ML successfully detects the conflicts and reduces the human force and effort significantly compared to the case of a passive robot that always follows human partner and a proactive robot that cannot resolve conflicts. In summary, in this dissertation, we propose pHRI approaches to adjust the contribution of the robot collaboratively manipulates an object with a human partner. Both approaches rely on the haptic data as inputs to the admittance controller to regulate the physical interaction between human and robot in an object co-manipulation task. Moreover, these approaches involve human intentions in their structures. In particular, in the first one, we introduce a rule-based estimation of human movement intentions and adapt the contribution of the robot to the manipulation task accordingly. On the other hand, in the second approach, since we consider a more complex scenario that involves a proactive robot, we take into account the dyadic intentions of both human and a robot in terms of the interaction behaviors that emerge between the partners during the co-manipulation task. We build a ML classifier that detects these behaviors and notifies the controller to adjust the role of the robot accordingly.