Learning of Decision Structures for Industrial Automation
- Reference number
- ID18-0096
- Start and end dates
- 190101-231231
- Amount granted
- 2 500 000 SEK
- Administrative organization
- KTH - Royal Institute of Technology
- Research area
- Information, Communication and Systems Technology
Summary
In this project, we will improve the ability of collaborative robots to autonomously learn decision structures such as behavior trees. It could be an important step towards “show and tell” like programming which would drastically reduce programming times and potentially spark an unprecedented revolution in automation. The project will consider several scenarios of gradually increasing complexity. Using a common interface module system for assembly, a variation of experiments will be tested. There are two main research areas that will be investigated in the project continuously. How to best determine a state model representation to describe the system and what methods are best for learning decision structures under different circumstances. The two areas are interdependent and must therefore be explored simultaneously. The main objective will thus be to develop a representation for robot tasks, that can be used both for automated learning of plans for new tasks, and in interaction with a human operator to significantly simplify the development of new robot programs.
Popular science description
The project will work to make robots smarter, working toward an end goal where eventually robots can be told and shown what to do, rather than being programmed. When every possible event does not have to be explicitly programmed, robots will also be able to work in more dynamic environments. This will make robotics more accessible and flexible. We will see robots used in many more areas where it is not possible today, increasing productivity and removing boring or dangerous work from humans.