NEAL: 2014-2016
- The main goal of the first two years was to get an
agent running in a closed loop virtual environment.
- We'd hoped to use the Neurorobotics virtal environment, but I
still haven't gotten that. So, I wrote my own in python.
- We figured out PyNN and went with Nest.
- We tried different neural models, but settled on IF_curr_exp neurons.
- We made a CABot1, which was open looped. That worked in
SpiNNaker.
- We made a CABot3, which was close looped. That worked in a version
of the SpiNNaker software (4 and 48 chip boards).
- It took NL commands, viewed the environment, followed plans, and
learned a really simple cognitive map (using the Tsodykys-Markram
synapse).
- We had hoped to make a radically improved version, but spent a lot
more time than expected rewritting software, and developing
neural and synapse models.
- We also got a persistent CA going in HICANN on the portal.
- We managed to get a neuro-cognitive model of categorisation (running
in our old java simulator). The paper is out for review.
- We did that all with standard neural and synapse models.
- We developed neural and synapse models in Nest and SpiNNaker.
- Ian did some nice stuff with plans, and is currently working on
learning plans.
