GPU-accelerated insect (and other!) brain simulations
The researchers from the University of Sussex at Capocaccia this year are part of an EPSRC-funded project called Brains-on-Board - now in its second year. Within this project we are developing insect-inspired models of vision, navigation etc to run in realtime on robotic platforms powered by hardware such as the NVIDIA Jetson TX1.
Here at Capocaccia we have two wheeled robots equipped with Jetson TX1s, panoramic cameras, IMUs and other sensors which we believe represent a very promising and flexible platform for embodied robotics research. While these robots can run TensorFlow and OpenCV, we have also developed an open source library called GeNN to simulate spiking neural networks using NVIDIA GPUs. The first aim of this workgroup is to help any interested users get up and running with GeNN and use it to accelerate the simulation of their own models.
Over the last year we have also implemented models of insect path integration and visual homing which can run on this robot platform. The second aim of this workgroup is to further develop these models - possible project areas include:
* At Capocaccia 2017 we developed an interface between GeNN and the DVS128 and used this to implement an optical flow algorithm - use this, rather than OpenCV, to provide a speed signal to the path integration model.
* Work on implementing these models on Neuromorphic hardware.
* Integrate simple spiking model of insect mushroom body into the current visual homing robot.
* The role of insects early visual systems in 'pre-processing' views of the world is somewhat mysterious - experiment with image processing (OpenCV) or machine learning (TensorFlow) algorithms to identify optimal computation.
Please see wiki page for links, downloads and more information!
|Tue, 24.04.2018||16:00 - 17:00||Lab|
|Wed, 25.04.2018||21:00 - 22:00||Lab|