Periodic Reporting for period 1 - CoLLaboratE (Co-production CeLL performing Human-Robot Collaborative AssEmbly)
Reporting period: 2018-10-01 to 2020-03-31
For CoLLaboratE to successfully realize its vision, a few scientific and technological objectives have been set throughout the project duration. These are listed in the following points:
1) To equip the robotic agents with basic collaboration skills easily adaptable to specific tasks
2) To develop a framework that enables non-experts teaching human-robot collaborative tasks from demonstration
3) The development of technologies that will enable autonomous assembly policy learning and policy improvement
4) To develop advanced safety strategies allowing effective human robot cooperation with no barriers and ergonomic performance monitoring
5) To develop techniques for controlling the production line while making optimal use of the resources by generating efficient production plans, employing reconfigurable hardware design, and utilising AGV’s with increased autonomy
6) To investigate the impact of Human-Robot Collaboration to the workers’ job satisfaction, as well as test easily applicable interventions in order to increase trust, satisfaction and performance
7) To validate CoLLaboratE system’s ability to facilitate genuine collaboration between robots and humans
- The use-cases were defined from a functional, layout and architectural point of view, to extract properly the main requirements. Focused interviews with the industrial end-users and a wider public were made. The joint analysis of these results furnished an initial list of main requirements and specifications on which the scientific developments should focus. The CoLLaboratE system architecture was analytically defined using the Volere methodology, facilitating the technological development towards the future integration.
- A questionnaire for the evaluation of the social aspects that affect the perception of the job quality and job satisfaction was prepared. A study about acceptance criteria was performed and experiments were planned to evaluate and assess job satisfaction and quality. Possible interventions that can be used to facilitate the HRC during the implementation of the use cases were identified and tested at a small case study.
- Different modalities to teach a robot assembly tasks by demonstration were developed, including visual cues, kinesthetic teaching and mobile app interfaces. The demonstrated task was further improved by autonomous exploration and learning, as well as learning induced by eventual human interaction during the exploitation.
- Novel collaborative skills were developed for efficient and safe interaction between the human and the robot during collaborative task execution, including novel methods for adaptive control of robotic manipulators during object handling, load sharing, compliant control schemes with dynamic obstacle avoidance and safety through novel designs of tactile sensors. Collaboration with AGVs is also achieved by recognizing gestures of the worker with novel skeleton tracking algorithms and translating them into actions to command the mobile robot.
- Methods were developed for efficient collaboration between human workers, production supervisors, robot manipulators and AGVs, including user-friendly graphical interfaces, novel human skeleton tracking algorithms for obstacle avoidance with robot manipulators and AGVs, ergonomic monitoring of the human and high level task planners for optimal resource allocation.
- The first steps were made towards integrating the CoLLaboratE modules, based on the defined modular architecture and the interfaces between the modules.
- A technology handbook has been prepared with the exploitable components of the project.
- A method for predicting the operator’s intention during a collaborative object transfer so that the robot can actively adapt to the task, by becoming proactive and minimizing the effort required by the human.
- A method to combine robot compliance and tracking accuracy, making the robot safe and adaptable to unknown task dynamics.
- Methods on robot perception and cognition that are versatile, and able to adapt to various tasks. More specifically, methods exploiting RGBD data were developed for both object detection and tracking, as well as human skeleton tracking.
- Extension of movement primitives to behavior primitives, developing active learning of Bayesian Gaussian Mixture Models as an enabler for multi-modal control and trajectory policy learning.
- A correct DMP formulation for encoding orientation trajectories for controlling a robot in the Cartesian space; this novel formulation is correctly reproducing the demonstrated trajectory without any oscillations that may appear when the dominant formulation is utilized.
- A method that utilizes virtual fixtures and DMP to assist the teaching and modification of a task, reducing the human’s cognitive and physical load.
- A framework that allows incremental refining of existing robot policy through kinesthetic guidance, to shorten the time necessary for the deployment of robot tasks by reusing existing similar policies.
- Two key technologies that lead to autonomous and semi-autonomous learning of robotic tasks. One using hierarchical reinforcement learning to learn the disassembly process and then to reverse the learned procedure and one for collaborative learning of exception strategies that help the robot autonomously resolve contingencies in the assembly process.
- An ergodic control algorithm for insertion-like tasks which requires stochastic exploration for task achievement. Ergodic control results in a natural exploration behavior to generate search patterns for such tasks.
- A human tracking algorithms that can be used to construct constraints to increase the safety for human-robot collaboration.
The expected potential impact from the project, as it has been identified by the potential end-users, is highlighted as follows:
- Introduction of human-robot collaboration on continuous production lines and tasks that were previously deemed not appropriate for robot automation
- Ergonomic improvement, particularly in intensive tasks performed manually such as riveting.
- Easy integration in production, drastically reducing the cost of programming.
- Versatility and adaptability, as the robot can be easily taught various collaborative activities.