IPAB Workshop - 21/09/2017

Speaker: Floyd Chitalu

Title: Lightweight Broad-Phase Collision Detection on the GPU

Abstract:

Broad-phase collision detection, as a highly dynamic task of proximity tests between interacting mesh geometry, is challenging in terms of parallelisation using GPUs.  As such, managing potentially divergent thread activity, even for a small number of complexly interacting objects, is difficult due to the overhead of work-flow divergence and potential for inefficient data access patterns which incur latencies that are detrimental to performance.  Additionally, the ability to adapt an algorithm for performance, which is ideal for the ever changing GPU platform, places further demands on implementation choices and trade-offs. In this presentation, I will give a brief overview about my recent work on a simple alternative method for simultaneously traversing multiple bounding-volume hierarchies for GPU based broad-phase collision detection that minimises issues of work-flow divergence and load-imbalance. I will describe how traversal is reformulated as an iterative and bulk-synchronous operation to guarantee a uniform mode of execution across threads by applying a topology-centred work expansion scheme to make use of the abundant parallelism and ensure simultaneous traversal of BVHs with different depths.

Speaker: John Pisokas

Title: The Neural Substrate of Path Integration in Insects

Abstract:

Understanding the neural circuits underlying insect behaviour has the potential to not only provide us with inspiration for solutions in robotics but also to elicit the principles of neural organisation, computation and function.

In this talk I will take you on a journey from the observed foraging behaviour to the neural circuits underlying this behaviour and I will show you where we stand and what problems we still have to address.