True Cobotics

Starting Point

Collaborative robots (Cobots) can interact with humans in a safe way. Multi-task and flexible, they are particularly adapted for small to medium production lots and customized production. They can also relieve human workers of tedious tasks and heavy work, revaluing their jobs. Correctly implemented, cobots promise higher productivity, avoiding offshoring in high-wage countries such as Switzerland.

The implementation of such systems is alas challenging, in particular for SMEs. The cobot must be integrated in a working system that satisfies safety regulations. In true Human-Robot Collaboration the workspace is shared; the environment varies with time, making programming an arduous task. An expert programmer is required which drastically reduces the advantages of cobotics, namely flexibility and low cost.

However, new developments in machine learning and teaching from demonstration offer ways to leverage the a priori knowledge on the task in order to develop smarter systems that can be more easily programmed.

Goals

A line worker shall be able to teach a cobot how to perform a task just as intuitively as he would teach a human colleague. A cobot shall be able to truly collaborate with a line worker to perform a task together. This will enable Swiss firms to become more efficient and remain globally competitive.

Approach

A collaborative robotic system can be decomposed in three parts: the robot, a workspace equipped with sensors and the human user. Each of these three entities needs to communicate with each other but have different languages, which makes the fusion of information arduous on the technical side and the communication difficult on the human-machine interaction side.

In this project, we develop a model of the environment through databases of the a priori knowledge of the task, allowing us to:

• develop a modular software architecture for the communication between different information systems (robot, cameras, microphones, etc.)
• exchange information between the system and the human in a natural way.

The databases are easily extendable by non-expert users, which ensures that the developed system is flexible and can be easily adapted to new situations.

Solution

A robotic solution has been developed that allows for intuitive human-robot interaction based on a common dictionary of actions and objects. This dictionary is easily extendable to new tasks in order to make the system flexible. In addition, the architecture is modular and can be used with different robots, different sensors and different programming languages.

Results

The flexibility of the architecture was demonstrated in three different use cases in industrial and service settings including three different robots:

• sensor packing
• drink distribution
• dish handling

The system developed allows people without any technical background to program a robot to perform variations of a well-defined task using intuitive interfaces. Adapting the system to completely new tasks and workspaces still requires trained engineers. Further development of the technology into a product and specific installations will be developed in the future with industrial partners. Interested partners are invited to get in touch with us.