To build this envisioned integrated robotic system, this project first designed a collaborative robot motion generation strategy that explores grasp freedoms. We designed a technique that allows multi-arm robots to change their grasps "in-the-air", similarly as a human would do, and adjust the gripper position in order to minimize the number of regrasps (making the interaction more fluid). This was then extended to also account for possible environment contacts (and contact points from the gripper and the robot arm). In this way, surfaces available in the environment could be explored to improve the stabilization of the shared object.
Secondly, we introduced freedoms in the workspace thus allowing the robot to deviate from the desired pose/trajectory within the region where the human is still comfortable. To this aim, we designed and evaluated different assessments for human comfort: peripersonal and muscular comfort and postural ergonomics.
Peripersonal comfort relates to the psychological concept of peripersonal space that accounts for the workspace where the human feels comfortable to perform a task. It is useful to improve human awareness, safety perception and minimize the perceived risk of robot intervention.
Muscular comfort accounts for the biomechanics response during the exchange of forces. To reason over human biomechanics perspective, we modelled the influence of external forces over different musculoskeletal models for the upper-limb. With this information, we built a predictive model capable of analyzing the human muscular response and also predict the kinematic configuration (position of the joints) during physical interaction with the robot.
Finally, we integrated established concepts for industrial ergonomics with muscular-informed comfort to introduce the comfortability concept. The comfortability allows assessment of comfort in a similar fashion to the way robot manipulability computes a quality index for manipulation tasks. In addition, we designed a method to quickly assess the comfortability distribution over human workspace. Our methodology based on precomputing relevant information about ergonomics and muscular capability considerably simplifies the comfortability distribution. This comfortability distribution thus allows a designer to observe regions in the human workspace (where the arm reaches) where the human is more likely to find a comfort configuration to interact.
The results obtained from combining the predictive power of the muscular-based model and the human-centred planning and control strategies demonstrate the success of this project and reveal the potential of the project to improve human-robot interaction performance, alleviate and reduce stress and lower incidence of musculoskeletal disorders (the largest cause of work-related injuries in many industrial countries). From human-robot experiments, we have found that the comfort-based planner is able to, on average, reduce the muscular load by 69.5% compared to a user-based selection of poses. This strategy is key to achieving intuitive and fluid human-robot interaction. Experiments were performed in collaboration with the research group from Dr Chakrabarty, a sensory-motor neurophysiology specialist from the University of Leeds.
Overview of results and dissemination
---
From a scientific perspective, this project obtained results have been published in three of the most important robotics conferences (IROS-18, Humanoids-18, Humanoids-19), and one major journal (AURO) in the robotics community. Additionally, there are other two journal publications under review focused on human-robot applications (ACM-THRI and RAL) and another under preparation.
Obtained results were also shared to the UK robotics community at the 3rd UK Robot Manipulation Workshop, the main robotics event in the UK. We organized the 2019 workshop which involved approximately 110 delegates from roughly 32 UK universities and industries nationwide. During this Workshop, we also organized demos to showcase our methods and potential advantages to fellow researchers and research teams from industry laboratories, e.g. Ocado, Amazon Robotics, Google Deepmind, and Dyson.
For the general community, this project also produced a tool for collaborative AI that allows the computation of a human comfort quality index distribution, the Rapid Human-Robot Manipulability Assessment. The RHuMAn draws from the predictive models obtained from this project to produce an efficient comfort-quality assessment that can be quickly tailored to specific tasks and purposes. The tool is available at the AI4EU European Project.
