The motivation for co-design stems from the core NCEI hypothesis that intelligence emerges from the dynamic interactions between an Embodied AI’s body, its environment, and its sensory-motor and computing capabilities. This contrasts sharply with the current State-of-the-Art (SOTA) in robotics, which typically separates the design process of a robot’s body and its autonomy stack, known as design compartmentalization. NCEI assumes that superior physical intelligence and agility requires a collective and cohesive treatment of the robot body and brain functions in a "closed-loop" fashion, analogous to evolution through natural selection.
The University of Oslo is a member of the NCEI centre, and works on implementing the "Body-and-Brain" Computational Co-design framework. The goal is to encode robot bodies in parametric terms and employ techniques such as evolutionary algorithms to computationally evolve them based on the fitness of their autonomy policies in navigation and manipulation tasks. We would like to explore methods like quality-diversity evolutionary algorithms in combination with deep reinforcement learning or bio-inspired control methods and massively parallel simulation to achieve these goals. In addition we would like to explore how some of these concepts could be applied on real-world robot platforms. A master’s project on this would typically either have a strong focus on algorithm development and simulation, or applying some algorithms in real-world robotic experiments.
A Master's project would normally involve the following components:
- Literature survey and writing essay
- Implementation of some experimental system, e.g. programming a variant of an evolutionary algorithm, designing test scenarios, or fitness functions
- Running experiments on the implemented approach. This may be using a physics simulation system, and/or on a real robot
- Analyzing the results and re-iterating on the steps above
- Writing the MSc thesis document