Assessing the potential of the rust fungus Puccinia spegazzinii as a classical biological control agent for the invasive weed Mikania micrantha in Papua New Guinea

The rust fungus Puccinia spegazzinii was introduced into Papua New Guinea (PNG) in 2008 as a classical biological control agent of the invasive weed Mikania micrantha (Asteraceae), following its earlier release in India, mainland China and Taiwan. Prior to implementing field releases in PNG, assessments were conducted to determine the most suitable rust pathotype for the country, potential for damage to non-target species, most efficient culturing method and potential impact to M. micrantha. The pathotype from eastern Ecuador was selected from the seven pathotypes tested, since all the plant populations evaluated from PNG were highly susceptible to it. None of the 11 plant species (representing eight families) tested to confirm host specificity showed symptoms of infection, supporting previous host range determination. A method of mass-producing inoculum of the rust fungus, using a simple technology which can be readily replicated in other countries, was developed. Comparative growth trials over one rust generation showed that M. micrantha plants infected with the rust generally had both lower growth rates and lower final dry weights, and produced fewer nodes than uninfected plants. There were significant correlations between the number of pustules and (a) the growth rate, (b) number of new nodes and (c) final total dry weight of single-stemmed plants placed in open sunlight and between the number of pustules and number of new nodes of multi-stemmed plants placed under cocoa trees. The trials suggest that field densities of M. micrantha could be reduced if the rust populations are sufficiently high. Crown Copyright (C) 2013 Published by Elsevier Inc. All rights reserved.

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.