IPAB workshop - Emmanuel Kahembwe/Paul Henderson/Tianqi Wei

Speaker: Emmanuel Kahembwe

Abstract: Joint actions between robots (and people) require coordination, in addition to traditional motion planning. Standard motion planning methods enforce coordination explicitly and reason about actions at the level of discrete start or end configurations with a trajectory connecting them. We present a concise encoding that combines motion planning and coordination. This encoding generalises the actions involved in joint tasks to continuous sets of trajectories in a parameterised manner. We efficiently learn this encoding and use it to control task execution.

Speaker: Paul Henderson

Abstract: Minimisation of discrete energies defined over factors is an important problem in computer vision, and a vast number of MAP inference algorithms have been proposed. We present a technique to automatically select the best inference algorithm for an input problem, by learning classifiers operating on problem features such as the factor structure and connectivity. On a diverse set of vision problems, this method selects an inference algorithm yielding labellings with an average of 96% of variables the same as the best available algorithm.

Speaker: Tianqi Wei

Abstract:Drosophila has a relatively simple neural system, but able to perform complex learning behaviours. To study learning behaviours of Drosophila larvae and test models of their neural circuits, 3 animats, which are a kind of robots mimicking animal behaviours, are being developed. The first is a soft pneumatic silicone robot that mimics larval peristalsis and tissue dynamic, of which a muscle is able to be controlled for simple motion currently. The second animat is a Lego wheeled robot mimicking navigation in ordure gradient, which is able to find a point with the strongest light in a light gradient. And the third is an analogue circuit wheeled robot with a control circuit mimicking motor neural circuit, of which the control circuit is still being simulated currently. They will be tested combine with Drosophila larval neural network simulation, and the results will be compared with the results of experiment on Drosophila larva.