Autonomous navigation in the (animal and the) machine


Information about my PhD project and the [neoRL] framework for autonomous navigation.

View My GitHub Profile

[ neoRL ] [ demonstration ] [ about the author ]

PhD — Autonomous Navigation

Doina Precup (DeepMind Montreal) and Harm van Seijen (Microsoft Research) found the neoRL to be an innovative contribution to the field of AI, accepting my PhD in computer science.

The full thesis is available here

Summary: neoRL networks

The neoRL approach to navigational autonomy is based on state representation in the mammalian brain, exploring neural representation of Euclidean space (NRES) for emulating online autonomous navigation in technology. The foundation of decomposing the prediction problem for RL is presented in Decomposing the Prediction Problem, demonstrating NRES-oriented RL (neoRL) for autonomous navigation.

Modern cognitive neuroscience further explores how reasoning itself can be a navigational challenge, after indications on how concepts activate NRES structures. My second article, presented at the Conference series in Artificial General Intelligence, explores how neoRL is capable of navigation across multiple parameter spaces, allowing for general and efficient navigation in high-dimensional space.

As presented at RLDM 2022, the neoRL framework is demonstrated to be capable of category II autonomy. Networks of neoRL nodes can establish deeper desires, where the output of one node is used as setpoint for another – giving a substantially improved navigational performance. The unreasonable efficiency of recurrent purposive neoRL networks is demonstrated under ``demonstrations’’, an effect that is explored further in my submission to RLDM-2022.


Decomposing the Prediction Problem

Navigating Conceptual Space; A new take on AGI

Towards neoRL networks; the emergence of purposive graphs

PhD thesis: Autonomous Navigation in (the Animal and) the Machine