University of Cambridge > > Robotics Seminar Series > Neuroevolutionary approaches to generating increasingly intelligent behaviours in virtual creatures / (simulated) robots

Neuroevolutionary approaches to generating increasingly intelligent behaviours in virtual creatures / (simulated) robots

Add to your list(s) Download to your calendar using vCal

If you have a question about this talk, please contact Amanda Prorok.

Dr Channon and his Evolutionary Systems group carry out research into the use of evolution to generate increasingly intelligent agents. Their work has included evolving deep neurocontrollers (deep neuroevolution) since 1996, using neural development (augmenting topologies) and generative encodings within that and other work also since 1996, and incorporating (initially static, hand-designed and later evolvable) convolutional neural networks with rectified linear units (ReLUs) since 2005. This talk will give an overview of some incremental approaches to the evolution of neural networks for the control of intelligent agents / virtual creatures / simulated robots. The focus here will be on ‘long-term’ evolution and approaches that generate increasingly intelligent behaviours. This part of the talk will conclude with a very brief overview of related work on open-ended evolution, in which the aim is to achieve ongoing adaptive novelty and ongoing growth of complexity.

The second part of the talk will focus on two specific examples of the incremental neuroevolution of virtual creatures / simulated robots. In the first example, a simple simulated quadruped is evolved to walk over obstacles (walls) of different heights. A range of different evolutionary complexification strategies are compared and ‘heterogeneous strategies’ are shown to overcome previous approaches’ shortcomings in relation to loss-of-gradient and over-fitting (analogous to catastrophic forgetting in neural systems). The second example demonstrates the incremental neuroevolution of reactive and deliberative 3D Agents. The 3D physically-based setting requires that a successful agent continually and deliberately adjust its gait, turning and other motor control over the many stages and sub-stages of these tasks, within its individual evaluation. Achieving such complex interplay between motor control and deliberative control, within a neuroevolutionary framework, is the focus of this work. The results demonstrate a variety of intricate lifelike behaviours being used, separately and in combination.

This talk is part of the Robotics Seminar Series series.

Tell a friend about this talk:

This talk is included in these lists:

Note that ex-directory lists are not shown.


© 2006-2023, University of Cambridge. Contact Us | Help and Documentation | Privacy and Publicity