I'm interested in using technology to enhance human cognitive function. My current research examines how to use intracranial EEG data generated as an animal navigates an environment to generate a map of that environment; the goal is to aid understanding of the properties of the internal maps encoded within our brains, and to provide a tool for testing to what extent different brain regions are neccesary for the construction and maintenence of these maps.

Simultaneous Localization and Mapping (SLAM) is the computational problem of constructing a map of an unknown environment while simultaneously keeping track of an agent's location within that environment. Several approximate solutions exist for solving this problem for robotic agents, including the extended Kalman filter, Covariance intersection, and GraphSLAM algorithms. These approaches are used employed for all manner of autonomously navigating robots – self driving cars, unmanned aerial vehicles, autonomous underwater vehicles, etc.  

Initial SLAM algorithms took camera input, as well as odometry readings, to perform SLAM (called visual-SLAM). Newer algorithms, however, have been developed which take a wide range of modalities as input - LiDAR-based SLAM algorithms, for example, uses LiDAR scanner readings as input, Acoustic SLAM uses audio signals as input, and wifi-SLAM uses the strengths of nearby wifi access points as sensor readings. In general, SLAM algorithms are highly adaptable to changes in input so long as input contains some location and odometry information. This research aims to develop a SLAM algorithm capable of operating on neural data gathered from rodents as they navigate a maze. The developed algorithm should be able to take this data as input and generate a reasonably accurate depiction of the maze dimensions, as well as a set of reasonably accurate location estimates for the rodent as it navigated the maze. 

Place cells are types of pyramidal neurons which typically exist in two regions of the hippocampus; the CA1 region, perhaps associated with self-identity, self-continuity, and self-awareness, and the CA3 region. Both of these regions are involved in the encoding of spatial representations and episodic memories. Place cells are characterised by the relationship between their firing rates and the location of the animal they exist within. For example, the firing rate of a place cell existing in the hippocampus of a rat exploring a maze may increase dramatically when that rat moves through the top left-hand corner of the maze. Cells whose firing patterns encode information about head direction and speed have also been discovered. Using neural data generated by these cells, location decoding has already been achieved, with highest decoding accuracy achieved by the Deepinsight framework.  

Deepinsight provides a general framework for neural decoding and has also been used for accurate decoding of head direction and speed. By utilizing this usage of Deepinsight, we would obtain a neural analogue of odometry readings which could be used for a developed SLAM algorithm. 

By developing a mapping algorithm capable of operating on neural data, we would gain a better understanding of the way representations of maps are stored within the brain. Moreover, this research could lead to biological agents being used to map environments, which may be advantageous to using robotic mapping agents in certain circumstances, for example for environments containing terrain which is difficult to traverse.