Simultaneous localisation and mapping (SLAM) is one of the core tasks of mobile autonomous robots. Powerful solutions to the SLAM problem exist but fast and low-power implementations are still being sought for. Since even simple animals navigate effortlessly in complex dynamically changing environments, taking inspiration from biology can lead to efficient solutions.
In this workgroup we will work on different parts of neuromorphic SLAM, inspired by hippocampal navigation system of rats. We will use two mixed-signal analog/digital neuromorphic devices — DYNAP-se and ROLLS, and two robotic vehicles — PUSHBOT and OMNIBOT, to implement and test the following architectures:
1. Path integration and odometry on chip
we will start with a system that can estimate heading-direction of the robot on chip, by integrating velocity commands in a spiking neural network. The architecture is described in the “HD_network_iscas18.pdf” located in the literature dropbox.
this system will be extended to include several rotation speeds
mapping between the speed commands and the shift network, which drives path integration, will be learned using sensory feedback and a match-mismatch circuit
the system will be extended to 2D path integration
2. Map formation
we will explore how a simple map of an environment can be learned using on-board plastic synapses of the neuromorphic device ROLLS. We will learn a collision map using the bumper sensor of the OMNIBOT robot (using synaptic potentiation) and explore how the map can be updated if the environment changes (using synaptic depression). We will also explore how collision predictions can be used to estimate accumulated errors of path integration.
3. Sequence learning
given a predefined map, we will study how a robot can navigate on a path according to a particular sequence of places. Through an implemented serial order architecture, the robot will learn a sequence of “places” on-chip and will navigate to these in the correct order during recall.
 Qiao N, Mostafa H, Corradi F, et al. A reconfigurable on-line learning spiking neuromorphic processor comprising 256 neurons and 128K synapses. Frontiers in Neuroscience. 2015;9:141. doi:10.3389/fnins.2015.00141
 Sandamirskaya, Yulia and Gregor Schoener (2010). “An embodied account of serial order: How instabilities drive sequence generation”. In: Neural Net- works 23.10, pp. 1164–1179. issn: 08936080. doi: 10.1016/j.neunet.2010. 07.012.
More literature about biologically inspired SLAM and models of hippocampal navigation