A bio-inspired computational model of motion detection

In many animal species it is essential to recognize approach predators from complex, dynamic visual scenes and timely initiate escape behaviors.
Locusts, for intance, provide ideal biological models that can be emulated in artificial systems, generating remarkably complex behaviors with low computational overhead. The aim of this work is to  develop a biologically-inspired comprehensive computacional model of complex motion detection that could be used to drive generic actuators for motor control. Specifically, the model is intended as a robust and effective collision detector for autonomous robot based applications. Innovation includes recent fidings predicting that the LGMD/DCMD neural circuitry in locusts encodes changes in the trajectories of approaching objects. The motivation is the number of applications in service tasks, ranging from the automotive industry, personal transportation devices for the physically impaired people (wheel chairs and walkers, for example), or motor control of prosthetic devices.

Goals:
The overall objective of this project is to develop a biologically-inspired, computationally simple model of complex motion detection that could be used  to drive generic actuators for robotics and other applications. The project will use a well-described insect collision detection system that consists of the locust Lobula Giant Movement Detector (LGMD) and it's postsynaptic partner, the Descending Contralateral Movement Detector (DCMD).
Therefore, the main goals of this project are:
- Teste a current, optimized model of the LGMD with simple and complicated visual stimuli and compare the obtained results to real LGMD data.
- Extract input-output algorithms based on the LGMD neuron responses to complex visual motion that relate dynamic visual stimulus parameters with modulation of the LGMD firing rate.
- Expand on a simplified model based on mean LGMD firing rates by incorporating input-output algorithms.
- Test the simplified LGMD model with different motion velocities and incresingly complex motion and compare the results to real LGMD responses. Finally, we aim to develop a more comprehensive computational model that reduces complexity and that could be used to program hardware for generic actuator control.

People involved in this research: