Colloquium Cognitive Systems
Spatial navigation in rodents with a transition scale-space, and an outlook to manipulation for robots
Dr. Nicolai Waniek (Bosch Center for Artificial Intelligence)
Abstract. How do animals represent space, and how can they use this to navigate through the world? The discovery of place, head, or -- more recently -- grid cells shed some light on this question and sparked significant research interest to answer it. However, we are far from an agreement about the involved computational mechanisms. In particular, the mesmerizing hexagonal responses and multi-scale representations of grid cells remain puzzling.
In this talk, I will introduce some of the existing models for grid cells, their strengths, but also their weaknesses. Following this, I will present a novel reward-free model that was developed using a top-down approach through Marr's suggested levels of analysis and with tools from computer science. Specifically, I propose that individual grid fields inform about local transitional knowledge to neighboring states of a suitable input space and that this knowledge of spatial relations is conveyed downstream to place cells. In turn, this can be used to plan navigational sequences. I will then present how to optimally accelerating such retrievals with well-known data structures from computer science, and how to apply this insight to neural responses. This forms the transition scale-space model for grid cells. In addition, the algorithmic approach also allows to derive a dynamical systems model of self-organizing grid cells.
Bio. After becoming a professional software developer in 2006, Nicolai Waniek decided to study computer science and, as a second field of study, biology at Ulm University. There, he did his diploma thesis under supervision of Prof. Dr. Heiko Neumann and graduated with a Diploma in Computer Science in 2012. Following, he moved to Technical University of Munich to the Neuroscientific System Theory, headed by Prof. Dr. Joerg Conradt, where he conducted research on spatial navigation in the rodent brain, and how to transfer the acquired insights to neuromorphic robotics. Moreover, he worked on event-based vision algorithms. He graduated from TU Munich with a Dr.rer.nat. Since May 2018 he is with the Bosch Center for Artificial Intelligence (BCAI), where he works in a project on manipulation robotics and leads a sub-team that conducts research in perception and pose estimation of highly symmetric objects with reflective surfaces.