Automatic camera exposure control

It is commonplace to use digital video cameras in robotic applications. These cameras have built-in exposure control but they do not have any knowledge of the environment, the lens being used, the important areas of the image and do not always produce optimal image exposure. Therefore, it is desirable and often necessary to control the exposure off the camera. In this paper we present a scheme for exposure control which enables the user application to determine the area of interest. The proposed scheme introduces an intermediate transparent layer between the camera and the user application which combines the information from these for optimal exposure production. We present results from indoor and outdoor scenarios using directional and fish-eye lenses showing the performance and advantages of this framework.

Exploiting parallax in panoramic capture to construct light fields

We show that the parallax motion resulting from non-nodal rotation in panorama capture can be exploited for light field construction from commodity hardware. Automated panoramic image capture typically seeks to rotate a camera exactly about its nodal point, for which no parallax motion is observed. This can be difficult or impossible to achieve due to limitations of the mounting or optical systems, and consequently a wide range of captured panoramas suffer from parallax between images. We show that by capturing such imagery over a regular grid of camera poses, then appropriately transforming the captured imagery to a common parameterisation, a light field can be constructed. The resulting four-dimensional image encodes scene geometry as well as texture, allowing an increasingly rich range of light field processing techniques to be applied. Employing an Ocular Robotics REV25 camera pointing system, we demonstrate light field capture,refocusing and low-light image enhancement.

Is the sun too bright in Queensland? An approach to robust outdoor colour beacon detection

Using cameras onboard a robot for detecting a coloured stationary target outdoors is a difficult task. Apart from the complexity of separating the target from the background scenery over different ranges, there are also the inconsistencies with direct and reflected illumination from the sun,clouds, moving and stationary objects. They can vary both the illumination on the target and its colour as perceived by the camera. In this paper, we analyse the effect of environment conditions, range to target, camera settings and image processing on the reported colours of various targets. The analysis indicates the colour space and camera configuration that provide the most consistent colour values over varying environment conditions and ranges. This information is used to develop a detection system that provides range and bearing to detected targets. The system is evaluated over various lighting conditions from bright sunlight, shadows and overcast days and demonstrates robust performance. The accuracy of the system is compared against a laser beacon detector with preliminary results indicating it to be a valuable asset for long-range coloured target detection.

Design of a four-rotor aerial robot

The design and fabrication of a proto-type four-rotor vertical take-off and landing (VTOL) aerial robot for use as indoor experimental robotics platform is presented. The flyer is termed an X4-flyer. A development of the dynamic model of the system is presented and a pilot augmentation control design is proposed.

Online learning of autonomous helicopter control

This paper details the development of an online adaptive control system, designed to learn from the actions of an instructing pilot. Three learning architectures, single layer neural networks (SLNN), multi-layer neural networks (MLNN), and fuzzy associative memories (FAM) are considerd. Each method has been tested in simulation. While the SLNN and MLNN provided adequate control under some simulation conditions, the addition of pilot noise and pilot variation during simulation training caused these methods to fail.

Obstacle detection for a mining vehicle using a 2D laser

This paper discusses a number of key issues for the development of robust obstacle detection systems for autonomous mining vehicles. Strategies for obstacle detection are described and an overview of the state-of-the-art in obstacle detection for outdoor autonomous vehicles using lasers is presented, with their applicability to the mining environment noted. The development of an obstacle detection system for a mining vehicle is then detailed. This system uses a 2D laser scanner as the prime sensor and combines dead-reckoning data with laser data to create local terrain maps. The slope of the terrain maps is then used to detect potential obstacles.

System design for a small autonomous helicopter

The detailed system design of a small experimental autonomous helicopter is described. The system requires no ground-to-helicopter communications and hence all automation hardware is on-board the helicopter. All elements of the system are described including the control computer, the flight computer (the helicopter-to-control-computer interface), the sensors and the software. A number of critical implementation issues are also discussed.

Height estimation for an autonomous helicopter

Height is a critical variable for helicopter hover control. In this paper we discuss, and present experimental results for, two different height sensing techniques: ultrasonic and stereo imaging, which have complementary characteristics. Feature-based stereo is used which provides a basis for visual odometry and attitude estimation in the future.

Field and Service Robotics: Results of the 9th International Conference

The Field and Service Robotics (FSR) conference is a single track conference with a specific focus on field and service applications of robotics technology. The goal of FSR is to report and encourage the development of field and service robotics. These are non-factory robots, typically mobile, that must operate in complex and dynamic environments. Typical field robotics applications include mining, agriculture, building and construction, forestry, cargo handling and so on. Field robots may operate on the ground (of Earth or planets), under the ground, underwater, in the air or in space. Service robots are those that work closely with humans, importantly the elderly and sick, to help them with their lives. The first FSR conference was held in Canberra, Australia, in 1997. Since then the meeting has been held every 2 years in Asia, America, Europe and Australia. It has been held in Canberra, Australia (1997), Pittsburgh, USA (1999), Helsinki, Finland (2001), Mount Fuji, Japan (2003), Port Douglas, Australia (2005), Chamonix, France (2007), Cambridge, USA (2009), Sendai, Japan (2012) and most recently in Brisbane, Australia (2013). This year we had 54 submissions of which 36 were selected for oral presentation. The organisers would like to thank the international committee for their invaluable contribution in the review process ensuring the overall quality of contributions. The organising committee would also like to thank Ben Upcroft, Felipe Gonzalez and Aaron McFadyen for helping with the organisation and proceedings. and proceedings. The conference was sponsored by the Australian Robotics and Automation Association (ARAA), CSIRO, Queensland University of Technology (QUT), Defence Science and Technology Organisation Australia (DSTO) and the Rio Tinto Centre for Mine Automation, University of Sydney.

Repeatable Condition-Invariant Visual Odometry for Sequence-Based Place Recognition

This paper describes a vision-only system for place recognition in environments that are tra- versed at different times of day, when chang- ing conditions drastically affect visual appear- ance, and at different speeds, where places aren’t visited at a consistent linear rate. The ma- jor contribution is the removal of wheel-based odometry from the previously presented algo- rithm (SMART), allowing the technique to op- erate on any camera-based device; in our case a mobile phone. While we show that the di- rect application of visual odometry to our night- time datasets does not achieve a level of perfor- mance typically needed, the VO requirements of SMART are orthogonal to typical usage: firstly only the magnitude of the velocity is required, and secondly the calculated velocity signal only needs to be repeatable in any one part of the environment over day and night cycles, but not necessarily globally consistent. Our results show that the smoothing effect of motion constraints is highly beneficial for achieving a locally consis- tent, lighting-independent velocity estimate. We also show that the advantage of our patch-based technique used previously for frame recogni- tion, surprisingly, does not transfer to VO, where SIFT demonstrates equally good performance. Nevertheless, we present the SMART system us- ing only vision, which performs sequence-base place recognition in extreme low-light condi- tions where standard 6-DOF VO fails and that improves place recognition performance over odometry-less benchmarks, approaching that of wheel odometry.

Reasoning about natural language phrases for semantic goal driven exploration

This paper presents a symbolic navigation system that uses spatial language descriptions to inform goal-directed exploration in unfamiliar office environments. An abstract map is created from a collection of natural language phrases describing the spatial layout of the environment. The spatial representation in the abstract map is controlled by a constraint based interpretation of each natural language phrase. In goal-directed exploration of an unseen office environment, the robot links the information in the abstract map to observed symbolic information and its grounded world representation. This paper demonstrates the ability of the system, in both simulated and real-world trials, to efficiently find target rooms in environments that it has never been to previously. In three unexplored environments, it is shown that on average the system travels only 8.42% further than the optimal path when using only natural language phrases to complete navigation tasks.

Distributed Robotic Vision as a service

The vision sense of standalone robots is limited by line of sight and onboard camera capabilities, but processing video from remote cameras puts a high computational burden on robots. This paper describes the Distributed Robotic Vision Service, DRVS, which implements an on-demand distributed visual object detection service. Robots specify visual information requirements in terms of regions of interest and object detection algorithms. DRVS dynamically distributes the object detection computation to remote vision systems with processing capabilities, and the robots receive high-level object detection information. DRVS relieves robots of managing sensor discovery and reduces data transmission compared to image sharing models of distributed vision. Navigating a sensorless robot from remote vision systems is demonstrated in simulation as a proof of concept.

Enhancing human action recognition with region proposals

Deep convolutional network models have dominated recent work in human action recognition as well as image classification. However, these methods are often unduly influenced by the image background, learning and exploiting the presence of cues in typical computer vision datasets. For unbiased robotics applications, the degree of variation and novelty in action backgrounds is far greater than in computer vision datasets. To address this challenge, we propose an “action region proposal” method that, informed by optical flow, extracts image regions likely to contain actions for input into the network both during training and testing. In a range of experiments, we demonstrate that manually segmenting the background is not enough; but through active action region proposals during training and testing, state-of-the-art or better performance can be achieved on individual spatial and temporal video components. Finally, we show by focusing attention through action region proposals, we can further improve upon the existing state-of-the-art in spatio-temporally fused action recognition performance.

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.

Characterisation of the Delphi Electronically Scanning Radar for robotics applications

Mm-wave radars have an important role to play in field robotics for applications that require reliable perception in challenging environmental conditions. This paper presents an experimental characterisation of the Delphi Electronically Scanning Radar (ESR) for mobile robotics applications. The performance of the sensor is evaluated in terms of detection ability and accuracy, for varying factors including: sensor temperature, time, target’s position, speed, shape and material. We also evaluate the sensor’s target separability performance.