Flexible cognitive robots for task completion in dynamic environments
Our aim is to develop novel human cognition inspired algorithms that can understand how humans complete tasks while dealing with variations in the environment as well as how they transfer learnt skills between various previously learnt tasks to solve new problems.
Background
Current collaborative robots (cobots) are programmed by kinaesthetic teaching which requires humans to guide the robot end effectors to do a task. The above technique often works well in environments that are rigid and well structured.
However, in situations where the task is often varying and the environment can change, these strategies fail. Furthermore, one of the paradigms that have been created by the machine learning community is the Reinforcement Learning paradigm.
Though state of the art reinforcement learning algorithms have given impressive results, they often have shortcomings when the variation in the environment or task increases.
On the other hand, natural cognition in humans provide us with a lot of food for thought in creating flexible, resilient and robust artificial intelligent systems. We take a user-inspired approach that works in close collaboration with stakeholders in industry and other sectors.
Our work revolves around
- Taking inspiration from both biology and psychology to design novel algorithms.
- Developing and deploying those novel algorithms on collaborative robots (Cobots)
- Validating the algorithms on various use cases defined by stakeholders.