IPAB Workshop - 17/05/2018

Speaker: Iordanis Chatzinikolaidis

Title: Dynamic footstep planning for legged robots using trajectory optimization

Abstract: Footstep planning for legged robots in arbitrary environments is a challenging task. The most successful approaches rely on very simplified models and assume level terrain to compute commands quickly. But these factors mean that the robots operate in a restricted domain of their capabilities. As a result, they are not reaching their movement potential when compared to the rich locomotion and balance skills which are present in nature. Recent advances in computational power and optimization allow us to expand the scope of what we can achieve by using more complex models and more challenging situations. In this talk we present an overview of the latest approaches, together with our formulation for tackling the problem. Finally, we will present some of our preliminary results.

Speaker: Christian Rauch

Title: Visual Articulated Tracking in the Presence of Occlusions

Abstract: Estimating and tracking the full state of an articulated object based on visual sensing enables many potential applications for robotic grasping tasks or human behaviour recognition. Current approaches to this problem use model-fitting and/or discriminative methods with a known model of the articulated object. However, these approaches are prone to failure when additional visual distractions, such as created by clutter or unmodelled objects, are introduced into the scene.

We present a tracking approach that combines model-fitting with discriminative information from pixel-wise segmentation to improve the robustness of model-fitting approaches. Further, we present a data augmentation strategy that introduces random occlusions during segmentation training to make the segmentation robust to visual distractions. This enables us to track a robot manipulator when it is occluded by or grasping a previously unseen object.

Speaker: Febrian Rachmadi

Title: Limited One-time Sampling Irregularity Age Map (LOTS-IAM): Automatic Unsupervised Detection of Brain White Matter Abnormalities in Structural Magnetic Resonance Images

Abstract: Supervised machine learning algorithms such as support vector machine (SVM), random forest (RF) and deep neural networks of CNN, uNet and uResNet are the most commonly used machine learning algorithms for automatic hyperintensities segmentation. However, all supervised methods are highly dependent on manual labels produced by experts (i.e. physicians) for training process. Furthermore, the quality of the label itself is dependent on and varies according to expert's skill and opinion, which rises questions about reproducibility in different sets of data. Unsupervised machine learning algorithms, such as Lesion Segmentation Tool toolbox (LST-LGA) and Lesion-TOADS which have been developed, tested in many studies and publicly available, do not need manual labels to work can eliminate the aforementioned dependency. Unfortunately, their performance is very limited compared to that from the supervised ones. We propose a novel approach of detecting and segmenting white matter abnormalities, named limited one-time sampling irregularity age map (LOTS-IAM), which outperforms the state-of-the-art unsupervised method of LST-LGA and competes with the state-of-the-art supervised method of deep neural networks algorithms.