Neuromorphic robot controllers: why, when, how?

Short abstract: In this discussion group we will talk about the place for neuromorphic controllers in the overall landscape of driving and table-top robots, discuss the challenges of integrating neuromorphic and robotic hardware, and set the goals for future research towards neuromorphic robots.

Login to become a member send


No timetable published yet.


Neuromorphic robot controllers

Neuromorphic hardware is particularly promising for robotic applications, in which robots have to produce behavior based on their own sensory information (perception) and previous experience (learning).  Neural networks have recently delivered impressive solutions to the problems of perception and learning in real world. These systems rely on fast processes of a large number of inputs in parallel, and the neuromorphic hardware offers an ideal substrate to process this kind of information in real time and with high energy efficiency, due to event-based and parallel nature of computation that this hardware realises.

However, a problem exists: most robots today were built with conventional computers in mind, and many manufacturers offer a high-level software interface for their robots, which allows to program them quickly using conventional methods, but don’t allow to access, in particular, motor control directly. Thus, the following questions arise:

  1. Are neuromorphic controllers for robots so beneficial indeed that it is worth changing the manufacturers’ policy and asking for direct control (which compromises safely and increases support efforts required)? Can we make a list of tasks, in which neuromorphic hardware (of which type?) brings clear advantages compared to conventional hardware, GPUs, and micro-controllers?
  2.  Can we collect a list of possible solutions for the interface between robotic sensors and motors and neuromorphic chips?What are the most promising approaches? Maybe we can make first steps towards a standard here.
  3. Which control strategies for robot actuators are compatible with neuromorphic hardware and can be realised efficiently? (attractor-based, VITE)
  4. How should a sensory pipeline look like in neuromorphic hardware for different sensor types?
  5.  Which “cognitive” or “behavioral” architectures are most promising for neuromorphic implementation?

In this Discussion group, we will brainstorm about these issues, hopefully getting some experts on board and otherwise working through examples in the literature. We will aim to arrive at a number of “white papers” or similar documents that sketch our suggested answers to the questions above. If time and manpower permits, we will try out different strategies with available robots and neuromorphic platforms to validate our ideas against reality of both the neuromorphic and the robotic hardware.

Suggested reading:

  1. Koziol, S.; Brink, S. & Hasler, J. A neuromorphic approach to path planning using a reconfigurable neuron array IC IEEE Transactions on Very Large Scale Integration (VLSI) Systems, 2014, 22, 2724-2737 
  2. Stewart, T. C.; Kleinhans, A.; Mundy, A. & Conradt, J. Serendipitous Offline Learning in a Neuromorphic Robot. Frontiers in Neurorobotics, 2016, 10, 1-11
  3. Krichmar, J. L. & Wagatsuma, H. Neuromorphic and brain-based robots. Frontiers in Artificial Intelligence and Applications, 2011, 233, 209-214 
  4. Perez-Pena, F.; Linares-Barranco, A. & Chicca, E. An approach to motor control for spike-based neuromorphic robotics. IEEE 2014 Biomedical Circuits and Systems Conference, BioCAS 2014 - Proceedings, 2014, 528-531
  5. Perez-Pena, F.; Morgado-Estevez, A.; Serrano-Gotarredona, T.; Gomez-Rodriguez, F.; Ferrer-Garcia, V.; Jimenez-Fernandez, A. & Linares-Barranco, A. Spike-based VITE control with dynamic vision sensor applied to an arm robot. Proceedings - IEEE International Symposium on Circuits and Systems, 2014, 463-466
  6. Corradi, F.; Moeys, D. P.; Corradi, F.; Kerr, E.; Vance, P.; Das, G.; Neil, D.; Kerr, D. & Delbrück, T. Steering a Predator Robot using a Mixed Frame / Event-Driven Convolutional Neural Network, Event-based Control, Communication, and Signal Processing (EBCCSP), 2016 Second International Conference on. IEEE,  2016
  7. Conradt, J.; Galluppi, F. & Stewart, T. C. Trainable sensorimotor mapping in a neuromorphic robot. Robotics and Autonomous Systems, Elsevier B.V., 2015, 71, 60-68
  8. Milde, M.; Blum, H.; Dietmüller, A; Sumislawska, D.; Conradt, J.; Indiveri, G. & Sandamirskaya, Y. Obstacle avoidance and target acquisition for robot navigation using a mixed signal analog/digital neuromorphic processing system. Frontiers in Neurorobotics, under review


Yulia Sandamirskaya


Eldad Assa
R B Benosman
Juan Pedro Dominguez-Morales
Emec Ercelik
Daniel Gutierrez-Galan
Xu He
Dongchen Liang
Fernando Perez-Peña
Sahana Prasanna
Matteo Ragni
Ole Richter
Jacopo Tani