This paper concerns the development of a force feedback enhanced teleoperation system for outdoor robotic vehicles navigating in rough terrain where true-colour 3D virtual world models of the working environment, created from laser and colour image scans collected offline, can be explored by walk-throughs both before and during the robot navigation mission itself. In other words, the physical mission planned can be partially rehearsed in cyberspace. Further, during a mission, the location and orientation of the vehicle are continually determined and global collision-free paths to selected goal locations made available as advice to the operator, who can follow or ignore such advice at will. Live (real-time) 3D laser range data also provides an up-to-date scan of the volume immediately surrounding the vehicle as it moves so that dynamic obstacles can be avoided. Local terrain-roughness is taken into account in the provision of local collision-free paths, the sub-goals of which, are operator determined. This live range data is matched with the pre-scanned range data to calculate the accurate robot vehicle localisation (position and orientation) which is provided continuously during the navigation mission. A force feedback 3D joystick reflects terrain roughness as a vibration in one axis and the other two axes are used to provide a 2D force to attract the operator towards following the local optimal collision-free path, but this attraction can be easily overridden by the operator. The instrumentation and methodologies used are presented, together with some preliminary experimental results.

This paper concerns the navigation of a physical robot in real natural environments which have been previously scanned in considerable (3D and colour image) detail so as to permit virtual exploration by cyber navigation prior to mission replication in the real world. An on-board high speed 3D laser scanner is used to localize the robot (determine its position and orientation) in its working environment by applying scan matching against the model data previously collected.

Visual interpretation of natural pointing gestures is essential in a human robot interaction scenario. Both hands and head are involved in pointing behaviors. Given the color images acquired by a web camera and the depth data by a TOF range camera, we perform visual tracking of the head and hands in 3D space. We investigate both the Head-Finger Line (HFL) and the forearm orientation as the estimation of the pointing direction. HFL is determined by 3D positions of the face and finger tip. Forearm direction is calculated using the PCA method in the RANSAC framework. Their performances are evaluated and compared in the experiments. Face direction and eye gaze orientation provide important cues regarding where the person's attention is during a pointing operation, which is proven helpful for eliminating some false estimations in our experiment.

This paper describes a topological SLAM system using a purely vision-based approach. This robot utilizes a GPU-based omnidirectional catadioptric stereovision system to perceive and plan its path in the environment. Subsequently, the omnidirectional images generated are used to incrementally build a database of image signatures based on the standard 2D Haar Wavelet decomposition. In order to maintain a globally consistent topological map, a relaxation algorithm, which requires local metric information between nodes, is employed each time the appearance-based localization system revisits an existing node in the topological map. The relative transformation of the current position of the robot with respect to the actual position of the matched node is recovered by using a least squares estimation of the transformation parameters of two 3D point patterns generated by the stereovision system. In addition, local metric information is obtained by using the proposed visual odometry system which combines distance measurements calculated by using optical flow techniques which estimates the movement of a web camera relative to the ground being observed and bearing estimates from the omnidirectional catadioptric vision system. Experiments were conducted in a variety of environments ranging from indoor to outdoor environments which demonstrate the feasibility of this approach.

This paper describes a probabilistic online map merging system for a single mobile robot. It performs intermittent exploration by fusing laser scan matching and omnidirectional vision. Moreover, it can also be adapted to a multi-robot system for large scale environments. Map merging is achieved by means of a probabilistic Haar-based place recognition system using omnidirectional images and is capable of discriminating new and previously visited locations in the current or previously collected maps. This dramatically reduces the search space for laser scan matching. The combination of laser range finding and omnidirectional vision is very attractive because they reinforce one another when there is sufficient structure and visual information in the environment. In other cases, they complement one another, leading to improved robustness of the system. This is the first system to combine a probabilistic Haar-based place recognition system using omnidirectional images with laser range finding to merge maps. The proposed system is also algorithmically simple, efficient and does not require any offline processing. Experimental results of the approach clearly illustrate that the proposed system can perform both online map merging and exploration robustly using a single robot configuration in a real indoor lab environment.

This paper outlines the framework of a complex system to demonstrate multimodal spatial and transactional intelligence in a robot which supports aged, frail, or otherwise disabled people in a domestic assistive technology context. The intention is that the robot be able to navigate around a known multi-room environment along optimal, collision-free paths in search and retrieval of requested objects such as spectacles, books etc. and must also be capable of tracking and following humans and of reminding them of times for meals, medication etc. and to lead disoriented subjects to their meal place at appropriate times and even dispense medication, if necessary. The modes of communication interchanges with the supported human include spoken speech and gestures (including eye gaze direction) within the context of situational analysis which accommodates recent history, temporal factors and individual user behavioural models. This paper provides an overview of an ambitious research project in its early stages but where many components have already been refined through earlier work.

This paper demonstrates how a robot can perform global localisation using a panoramic mirror in conjunction with a rich 3D model of the environment and a particle filter for localisation. The 3D model is created using Riegl Z420i laser range scans taken at various positions in the environment and registered together. An appearance-based map is created from this 3D model by sampling from it at various poses. For feature extraction we use Haar-wavelets as a global scene signature. In our experiments, the robot is able to localise on average within 1 metre of the true position and 5 degrees for orientation, in an environment measuring about 50times50 metres.

This paper demonstrates how robot navigation in a cyber world created by laser/image scanning a small portion of the real world can be replicated in the real world by a physical robot. A new paradigm for robot navigation is thus created.

Odour localisation in an enclosed area is difficult due to the formation of sectors of circulating airflow. Well-defined plumes do not exist, and reactive plume following may not be possible. Odour localisation has been partially achieved in this environment by using knowledge of airflow, and a search that relies on chemical sensing and reasoning. However the results are not specific, with the odour source only restricted to a broad area. This paper presents a solution to the problem by introducing a second search stage using visual sensing. It therefore comprises a bi-modal, two-stage search, with each stage exploiting complementary sensing modalities. This paper presents details of the method and experimental results

This paper describes a fusion of panoramic vision and laser range data to track multiple persons simultaneously from a stationary robot. Particle filters are used to track people in the plane of the laser and a mixture of Gaussians background subtraction algorithm is used to maintain a colour model for each person being tracked. Colour information is used to recognize lost targets that have reentered the scene

In this paper, a promising approach for visual target tracking is presented. A mobile robot should keep an unpredictably moving target in a cluttered environment within its field of view. The objective is mainly to plan online the shortest tracking path that offers a continuous monitoring of the target. A distance transform based Gaussian distribution for obstacle cluttered space is used to build a probabilistic model for the target future location. It identifies the target potential escapes out of the robot view. Accordingly, the robot navigates to an appropriate observing point that is calculated and updated continuously as the target moves. The approach has been tested on simulated static and dynamic environments. A number of test cases are presented

We describe a hybrid visuo-spatial topological approach in which global visual localization is triggered at junctions detected in a novel egocentric spatial approximation of the generalized Voronoi diagram. Transited vertices of this GVD are shown empirically to be useful and robust topological features. Visual vertex discrimination is performed without spatial cues, ameliorating junction ambiguity. Several large-scale trials are reported demonstrating fully autonomous exploration and mapping in mixed indoor/outdoor environments

Using artificial evolution successfully to design behaviours of multiple robot systems has been reported in recent years. Most of such reports are focused on the design of low level controllers. Design of high level team coordination strategies is rarely covered perhaps because the design of an appropriate chromosome representation for a complex multi-agent system is not an easy task. In this paper we propose that by treating the action decisions of every team member as a supervised ranking problem, the chromosome design problem can be solved systematically.We have tested this approach by dynamically solving the problems in the Solomon's benchmark of Vehicle Routing Problem with Time Windows [1]. Experiments show that our approach can create some simple behaviours which, whilst not optimal, are robust and above average in quality.

This paper describes an approach for solving a visibility-based covert path planning problem. It is an extension to our research on covert robotics which aims to create robots with the ability to achieve different covert navigation missions. A promising method is presented that allows a mobile robot in a complex, initially unknown environment to discover a path to a nominated destination with minimum exposure to sentries' at known locations within the same environment. The sentries' locations are known initially, suspected, or discovered during navigation. The presented method estimates a cost value at each non-obstacle location that represents a risk of being visible, given the accumulated map knowledge. A global covert path is generated by propagating the estimated cost values using the distance transform algorithm. The approach has been evaluated on both simulated and real environments. A number of test cases are presented; each shows a generated path with considerable covertness compared to a short path (but perhaps risky) to the same destination.

In this paper, we present a new hybrid method of performing eye-to-hand coordination and manipulation to produce a working robot named COERSU. The method is an optimized combination of two neuro-fuzzy approaches developed by the authors: direct fuzzy servoing and fuzzy correction. The fuzzy methods are tuned by an adaptive neuro-fuzzy inference system (ANFIS). On the whole, a genetic tuner and two neuro-fuzzy networks contribute to find the final optimum position of the robotic tooltip in order to grasp the target. Experimental results from COERSU in a table-top scenario to manipulate some soft objects (e.g. fruit/egg) also validate the method.

Designing control processes for mobile robot teams is a very difficult task. Consequently, methods for the teams to learn or adapt on their own are highly desirable. Genetic algorithms (GAs), which is a class of techniques inspired by biological evolution, is a mainstream method to accomplish this self-adaptation. However, a major problem of the genetic algorithm approach is the brittleness of the evolved solution(s). The solutions provided by GAs can only work properly in those environments closely resembling the training environment. A slight change in the environment can render a very poor performance. In short, the solutions lack robustness/generality and scalability. This paper discusses a design methodology to obtain robust solutions with the GA approach. The effectiveness of this methodology is demonstrated by an example. We have designed a mobile robot team with stable performance in a variety of environments.

Here, a robust tuner is presented that has been developed for support of early visual understanding by a robotic humanoid, sg1, under development. One of the achieved goals is to speed up the process of real-time segmentation by eliminating any tuning sessions from the online process and carrying them out offline. Effectiveness values (credits) assigned to stereo-based range findings and colour components (red, green, blue) are some of the tuned parameters. However, more than nine parameters (such as: stereo range finder search area, minimum and maximum size of regions, thresholds...) are tuned using a genetic algorithm. A novel idea for automatic evaluation of the region-edge segmentation has been applied as well.

Research projects about evolution of agents in a cellular world are not new topics in the artificial life (AL) fields. However, most of the studies focus on those fundamental, social behaviours like energy preservation, pattern formation or leader following etc. This paper presents experiments about applications of genetic algorithms (GAs) to an empirical multiple robot cooperative task: unknown environment exploration. These experiments investigate the effectiveness of GAs for evolving behaviours of individual swarm members that constitute good collective results. They try to answer the questions of (i) Can GAs find such behaviours, or, do such behaviours exist? (ii) Are these behaviours sensitive to environmental changes?.

This research concerns the development of a binocular computer vision system for a robotic humanoid using interactive hand/eye coordination as a focus for skills teaching and capability demonstrations. The humanoid form aids natural transference of sophisticated human skills of perception, spatial reasoning, sensor guided manipulation and person-to-person communications (eye contact, body language, speech, touch etc.) to the robot and simultaneously stimulates considerations of human centred applications requiring such skills. This research builds on a humanoid project undertaken within the Intelligent Robotics Research Centre at Monash University and funded internally in 1999. At this early stage of the project much of the material presented in this paper is speculative and part of the planning for an extended piece of research over a number of years.

Details the design and construction of a full sized robotic humanoid whose body is made entirely from lightweight polyurethane plastic and whose total size and weight are that of a typical human adult. The design of this robot was conceived with research in the conventional human environment in mind, incorporating all aspects of human mobility and sensory capabilities. Plastic components for the robot are easily and quickly manufactured, and are easy and cost effective to replace if damaged or broken. The substantial saving in weight that is possible through the use of plastics significantly affects other aspects of the design including motor and cabling capacities as well as battery life.

A nonparametric method for clustering multidimensional data in O(nlogn) time is described. It is based on the shared near neighbours algorithm. It uses adaptive k-d trees combined with various other sophisticated data structures to significantly decrease the computational complexity of the original algorithm which was O(n/sup 2/). The algorithm is suitable for a wide range of data and capable of delineating clusters of varying shape, density, and homogeneity. A comprehensive set of results is presented.

This paper describes the integration of sensory data, path planning methodology and hydraulic motor power to create an autonomous, heavy duty, outdoor robotic tracked vehicle capable of hauling loads in working environments which can be initially unknown and time-varying. Potential applications include agricultural, forestry and surface mining operations as well as firefighting and search and rescue missions. Progress to date is reported along with development plans for the near future.

This paper describes the collection, editing and reproduction of 3D structures, i.e. solid replication, with potential applications in computer vision, rapid phototyping, prosthesis, wind-tunnel modelling and scientific visualisation. Robotic transport of materials and 'manufactured' components is central to the approach adopted. The resulting system is economically viable and has commercial potential.

A practical solution to path planning problem for a six-degree-of-freedom robotic manipulator with arbitary payload has been implemented. This problem in its most general form has been shown to be intractable, so most researchers in the field simplify the problem in some way. Often these simplifications make the resulting path planner of little practical use. In this paper an overview of a practical path planner is given. Heuristics are used to plan for the last three degrees of freedom of the robot and its payload, and the details of one of these heuristics is described. The use of heuristics makes the planner incomplete, but for all tasks programmed for the robot the planner has not failed to find a path.

This work considers the essential requirements for outdoor mobile robot navigation in the context of converting existing, commercially available, outdoor vehicles to provide the platform for such endeavours. Two particular conversions are presented as examples to illustrate the approach being championed and a number of practical details to achieve this end presented as a guide to others who may be interested in pursuing this area of research with limited financial resources.

In recent times a great deal of interest has been shown, amongst the computer vision and robotics research community, in the acquisition of range data for supporting scene analysis leading to remote (noncontact) determination of configurations and space filling extents of three-dimensional object assemblages. This paper surveys a variety of approaches to generalized range finding and presents a perspective on their applicability and shortcomings in the context of computer vision studies.

The requirements of three-dimensional scene analysis in support of vision driven robotic manipulation are such that direct range finding capabilities can greatly reduce the computational burden (time and complexity) for reliable determination of the placement and spatial extent of objects in the scene. This paper describes a laser time-of-flight range scanner capable of acquiring a reasonable rangepic of 64 × 64 spatial resolution in 4 s. However, at the current state of development, 100 samples/point are required to achieve a range accuracy of approxi-mately ±¿ cm, with a scan time of 40 s. Returned signal amplitude de-pendency is also apparent in the range determination, the dynamic range of intensity being considerable. Improved accuracy can be traded for tine. The vision/robotics laboratory environment within which the range scanner operates is briefly touched upon and some preliminary rangepic results presented.

A great number of techniques have been applied to the general problem of adaptive control. What began as a study of engineering adaptive control problems involving dynamics, system and measurement noise, monitoring, transduction, and on-line instrumentation seems to have moved towards learning theory and methodology research that uses a refined plant/environment model as a vehicle of demonstration. An attempt is made to bring together, order, and briefly discuss many contributions in this field, bridging the era of earlier engineering practice to more recent artificial intelligence speculation. Both unimodal and multimodal strategies are discussed, together with problems arising in nonstationary environmental situations where information conservation, update, and retrieval are of considerable importance. Methods discussed include gradient, correlation, random, stochastic automata, fuzzy automata, pattern recognition, and mixed strategies. A selected reference list is provided.

A probabilistic automaton based on competing evolutionary search processes is considered as an on-line real-time adaptive controller for complex plants with multimodal performance index structures that are modified under measurable but uncontrollable random environmental variations. The scheme described includes mechanisms of storage and retrieval to take advantage of control experience gained prior to any current situation.

A nonparametric clustering technique incorporating the concept of similarity based on the sharing of near neighbors is presented. In addition to being an essentially paraliel approach, the computational elegance of the method is such that the scheme is applicable to a wide class of practical problems involving large sample size and high dimensionality. No attempt is made to show how a priori problem knowledge can be introduced into the procedure.