In a factory environment, execution of polishing and screwing tasks, are frequently encountered and can be easily accomplished by human workers. Nevertheless, automating these tasks with robots require expert robotics engineering knowledge, robot-specific arrangements in the factory, and exact model of the environment. Through the integration of human-to-robot skill transfer techniques, these monotonous tasks can be autonomously handled by a mobile manipulator so that human workers can focus on more complicated tasks. To this end, this project aims to develop required technologies to create such a system that can be actively used in a real factory environment for actual scenarios. With the successful completion of the project, a crucial building block will be built for human-centred robotic automation, which is of importance in keeping the manufacturing jobs in EU.
- The robot will be able to physically interact with humans in a safe and sound manner through the use of a novel haptic interface. This haptic interface will allow the robot to act mechanically transparent when necessary. In doing so, human workers can effortlessly manipulate the robot end-effector just like a tool while training the robot.
- The robot will be able to learn certain tasks by demonstration; human workers with no engineering expertise will use the robot as a tool to perform the task. Then the robot will be able to acquire necessary skills for autonomous task execution. Mainly drilling and screwing tasks will be targeted as they are the most two frequent tasks in industrial manufacturing.
- The robot will adaptively work in human working environment with no special arrangements. It will be fully reconfigurable for different industrial tasks in different settings.
Phase I: System Integration:
we construct a robust and reliable mobile manipulation platform using off-the-shelf tools and products. The robot will be able to physically interact with humans in a safe and sound manner through the use of a novel haptic interface. This haptic interface will allow the robot to act mechanically transparent when necessary. Thus, human workers can effortlessly manipulate the robot end-effector just like a tool while training the robot.
Phase II: Human-to-robot skill transfer
The objective of Phase-2 (Human Skill Acquisition) is to synthesize controllers state-to-action mapping) to achieve skill acquisition from humans. Therefore, we collect an extensive amount of data as the robot is trained by a human concerning the target tasks. i.e., primarily polishing. Since the target tasks enforce the robot to physically interact with the environment, the dataset will include all possibilities in the factory domain. Once data collection process is finished, state to action mapping will be constructed using machine learning algorithms to obtain the control policy.
Fig.5: The proposed mobile manipulator
- A mobile manipulator that can effectively learn impedance critical tasks (e.g. drilling, screwing) from human with no engineering expertise (human workers) and execute subsequent autonomous tasks.
- A trainable robot which allow the human workers to effortlessly manipulate the robot end-effector just like a tool in the training.
- A non-stationary robot with a lightweight robot arm and a mobile base with Omni wheel steering that can move easily and safely in the factory environment (instead of a caged cell placement robot)
- A complete system that can be actively used in a real factory environment for actual scenarios.