Towards vision-based deep reinforcement learning for robotic motion control

This paper introduces a machine learning based system for controlling a robotic manipulator with visual perception only. The capability to autonomously learn robot controllers solely from raw-pixel images and without any prior knowledge of configuration is shown for the first time. We build upon the success of recent deep reinforcement learning and develop a system for learning target reaching with a three-joint robot manipulator using external visual observation. A Deep Q Network (DQN) was demonstrated to perform target reaching after training in simulation. Transferring the network to real hardware and real observation in a naive approach failed, but experiments show that the network works when replacing camera images with synthetic images.

Extending visual perception with haptic exploration for improved scene understanding

Scene understanding has been investigated from a mainly visual information point of view. Recently depth has been provided an extra wealth of information, allowing more geometric knowledge to fuse into scene understanding. Yet to form a holistic view, especially in robotic applications, one can create even more data by interacting with the world. In fact humans, when growing up, seem to heavily investigate the world around them by haptic exploration. We show an application of haptic exploration on a humanoid robot in cooperation with a learning method for object segmentation. The actions performed consecutively improve the segmentation of objects in the scene.

Characterisation of the impact of open biomass burning on urban air quality in Brisbane, Australia

Open biomass burning from wildfires and the prescribed burning of forests and farmland is a frequent occurrence in South-East Queensland (SEQ), Australia. This work reports on data collected from 10-30 September 2011, which covers the days before (10-14 September), during (15-20 September) and after (21-30 September) a period of biomass burning in SEQ. The aim of this project was to comprehensively quantify the impact of the biomass burning on air quality in Brisbane, the capital city of Queensland. A multi-parameter field measurement campaign was conducted and ambient air quality data from 13 monitoring stations across SEQ were analysed. During the burning period, the average concentrations of all measured pollutants increased (from 20% to 430%) compared to the non-burning period (both before and after burning), except for total xylenes. The average concentration of O3, NO2, SO2, benzene, formaldehyde, PM10, PM2.5 and visibility-reducing particles reached their highest levels for the year, which were up to 10 times higher than annual average levels, while PM10, PM2.5 and SO2 concentrations exceeded the WHO 24-hour guidelines and O3 concentration exceeded the WHO maximum 8-hour average threshold during the burning period. Overall spatial variations showed that all measured pollutants, with the exception of O3, were closer to spatial homogeneity during the burning compared to the non-burning period. In addition to the above, elevated concentrations of three biomass burning organic tracers (levoglucosan, mannosan and galactosan), together with the amount of non-refractory organic particles (PM1) and the average value of f60 (attributed to levoglucosan), reinforce that elevated pollutant concentration levels were due to emissions from open biomass burning events, 70% of which were prescribed burning events. This study, which is the first and most comprehensive of its kind in Australia, provides quantitative evidence of the significant impact of open biomass burning events, especially prescribed burning, on urban air quality. The current results provide a solid platform for more detailed health and modelling investigations in the future.