Increasing farm labour and input costs and the requirement for more orchard data are leading to rapid advances in technology to improve management systems in fruit production. The study aimed to (i) validate a sensorised platform to estimate fruit number, peel colour and blush coverage in pear orchards with different pear selections and tree architectures, (ii) establish relationships between fruit number, peel colour and canopy geometry features, and (iii) evaluate the platform for mapping and digitising orchard features. The study was carried out over two years in an experimental 'ANP-0131 ' orchard and in one year in two commercial pear orchards ('ANP-0131 ' and 'PremP009'). Predictions of fruit number and blush coverage were compared in traditional three-dimensional (3D) and modern high-density two-dimensional (2D) training systems. Overall, prediction errors for fruit number were < 6.5 % in all the training systems, but improved performance was achieved in vertical 2D configurations (% standard errors = 2.2 %). Fruit number and blush coverage per unit of leaf area were higher in 2D compared to 3D training systems. Blush coverage predictions in 'ANP-0131 ' were reliable (R-2 = 0.67, RMSE = 3.70 %). Accurate predictions of blush coverage classes were achieved by modifying sample variance. Fruit number and blush coverage were negatively affected by increasing canopy size. The platform proved useful for mapping and digitisation. Spatial heatmaps of orchard features provided a valuable visual aid to identify zones for priority interventions for peel colour enhancement.

This study investigated the adoption of a commercial, sensorised platform namely Green Atlas Cartographer, equipped with a network of proximal sensors (including cameras and LiDAR) and machine learning algorithms - to evaluate orchard-specific relationships between estimated tree geometry, fruit number, fruit clustering, fruit size and fruit colour in commercial apple and pear orchards. The study was conducted at three commercial apple and pear sites in the Goulburn Valley ( Victoria, Australia) during the 2021-2022 season. Estimations of canopy area, canopy height, canopy density and cross-sectional leaf area were generated using LiDAR technology. Fruit number, fruit clustering, fruit colour development and fruit diameter were estimated using a combination of machine vision and deep learning. The geo-referenced data points generated by Cartographer were grouped into spatial plots and statistics by plot were obtained. Correlation and principal component analyses of harvest crop parameters unveiled underlying relationships between tree geometry, productive performance and fruit quality attributes. The relationships obtained in this study can drive orchard design strategies and dictate management decisions so that trees can be standardised to consistently produce high-quality fruit over the lifespan of modern apple and pear orchards.

Automatic in-field fruit recognition techniques can be used to estimate fruit number, fruit size, fruit skin color, and yield in fruit crops. Fruit color and size represent two of the most important fruit quality parameters in stone fruit (Prunus sp.). This study aimed to evaluate the reliability of a commercial mobile platform, sensors, and artificial intelligence software system for fast estimates of fruit number, fruit size, and fruit skin color in peach (Prunus persica), nectarine (P. persica var. nucipersica), plum (Prunus salicina), and apricot (Prunus armeniaca), and to assess their spatial and temporal variability. An initial calibration was needed to obtain estimates of absolute fruit number per tree and a forecasted yield. However, the technology can also be used to produce fast relative density maps in stone fruit orchards. Fruit number prediction accuracy was ≥90% in all the crops and training systems under study. Overall, predictions of fruit number in two-dimensional training systems were slightly more accurate. Estimates of fruit diameter (FD) and color did not need an initial calibration. The FD predictions had percent standard errors <10% and root mean square error <5 mm under different training systems, row spacing, crops, and fruit position within the canopy. Hue angle, a color attribute previously associated with fruit maturity in peach and nectarine, was the color attribute that was best predicted by the mobile platform. A new color parameter—color development index (CDI), ranging from 0 to 1—was derived from hue angle. The adoption of CDI, which represents the color progression or distance from green, improved the interpretation of color measurements by end-users as opposed to hue angle and generated more robust color estimations in fruit that turn purple when ripe, such as dark plum. Spatial maps of fruit number, FD, and CDI obtained with the mobile platform can be used to inform orchard decisions such as thinning, pruning, spraying, and harvest timing. The importance and application of crop yield and fruit quality real-time assessments and forecasts are discussed.

This paper evaluates the efficacy of a machine learning approach to data fusion using convolved multi-output Gaussian processes in the context of geological resource modeling. It empirically demonstrates that information integration across multiple information sources leads to superior estimates of all the quantities being modeled, compared to modeling them individually. Convolved multi-output Gaussian processes provide a powerful approach for simultaneous modeling of multiple quantities of interest while taking correlations between these quantities into consideration. Experiments are performed on large scale data taken from a mining context.

Modern horticulture is undergoing a rapid change with the introduction of new predictive technologies that help maximise the automation of orchard management practices. This study aimed to calibrate and validate a commercial sensorised mobile platform for the prediction of flower cluster number, fruit number and yield, tree geometry in 'ANABP-01' apples. In addition, this work (i) modelled the relationships between tree geometry and light interception, and (ii) determined the effects of light interception, rootstock and row orientation on flower cluster number, crop load, yield and tree geometry. Results showed that predictions were very accurate after initial calibration. Flower cluster detections had an error (RMSE) of similar to 5 clusters / image. Fruit number and yield predictions needed independent calibration across rootstocks but errors after validation on a separate dataset were small (RMSE = 5 fruit / tree, and RMSE = 1 kg / fruit, for fruit number and yield, respectively). Orchard errors for fruit number and yield estimations were lower than 5 %. Canopy area, canopy density and canopy cross-sectional leaf area (CSLA) were all linearly related with effective area of shade (EAS, integrated daily canopy light interception) but CSLA had the most robust and stable relationship with intercepted light. Increasing CSLA led to higher flower cluster number, fruit number and yield. Row orientations and rootstocks significantly affected productive performance, tree size and geometry and light interception. The orchard heatmaps generated after data validation proved very useful to support orchard management decisions. Overall, the predictive technology demonstrated to be a valid tool to combine accurate estimates of several important fruit crop parameters (i.e. flower cluster number, fruit number, yield, tree size and geometry, and light interception) in a single platform.

This paper proposes an analytical framework for modelling resource contention in multi-robot systems, where the travel times and task durations are uncertain. It uses several approximation methods to quickly and accurately calculate the probability distributions describing the times at which the tasks start and finish. Specific contributions include exact and fast approximation methods for calculating the probability of a set of independent normally distributed random events occurring in a given order, a method for calculating the most likely and n-th most likely orders of occurrence for a set of independent normally distributed random events that have equal standard deviations, and a method for approximating the conditional probability distributions of the events given a specific order of the events. The complete framework is shown to be faster than a Monte Carlo approach for the same accuracy in two multi-robot task allocation problems. In addition, the importance of incorporating uncertainty is demonstrated through a comparison with a deterministic method. This is a general framework that is agnostic to the optimisation method and objective function used, and is applicable to a wide range of problems.

The weekly maintenance schedule specifies when maintenance activities should be performed on the equipment, taking into account the availability of workers and maintenance bays, and other operational constraints. The current approach to generating this schedule is labour intensive and requires coordination between the maintenance schedulers and operations staff to minimise its impact on the operation of the mine. This paper presents methods for automatically generating this schedule from the list of maintenance tasks to be performed, the availability roster of the maintenance staff, and time windows in which each piece of equipment is available for maintenance. Both Mixed-Integer Linear Programming (MILP) and genetic algorithms are evaluated, with the genetic algorithm shown to significantly outperform the MILP. Two fitness functions for the genetic algorithm are also examined, with a linear fitness function outperforming an inverse fitness function by up to 5% for the same calculation time. The genetic algorithm approach is computationally fast, allowing the schedule to be rapidly recalculated in response to unexpected delays and breakdowns.

This paper develops an analytical method of truncating inequality constrained Gaussian distributed variables where the constraints are themselves described by Gaussian distributions. Existing truncation methods either assume hard constraints, or use numerical methods to handle uncertain constraints. The proposed approach introduces moment-based Gaussian approximations of the truncated distribution. This method can be applied to numerous problems, with the motivating problem being Kalman filtering with uncertain constraints. In a simulation example, the developed method is shown to outperform unconstrained Kalman filtering by over 40% and hard-constrained Kalman filtering by over 17%. (C) 2016 Elsevier B.V. All rights reserved.

This chapter presents an overview of the state of the art in mining robotics, from surface to underground applications, and beyond. Mining is the practice of extracting resources for utilitarian purposes. Today, the international business of mining is a heavily mechanized industry that exploits the use of large diesel and electric equipment. These machines must operate in harsh, dynamic, and uncertain environments such as, for example, in the high arctic, in extreme desert climates, and in deep underground tunnel networks where it can be very hot and humid. Applications of robotics in mining are broad and include robotic dozing, excavation, and haulage, robotic mapping and surveying, as well as robotic drilling and explosives handling. This chapter describes how many of these applications involve unique technical challenges for field roboticists. However, there are compelling reasons to advance the discipline of mining robotics, which include not only a desire on the part of miners to improve productivity, safety, and lower costs, but also out of a need to meet product demands by accessing orebodies situated in increasingly challenging conditions.

This paper describes an application of adaptive sampling to geology modeling with a view of improving the operational cost and efficiency in certain surface mining applications. The objectives are to minimize the number of blast holes drilled into, and the accidental penetrations of, the geological boundary of interest. These objectives are driven by economic considerations as the cost is, firstly, directly proportional to the number of holes drilled and secondly, related to the efficiency of target material recovery associated with excavation and blast damage. The problem formulation is therefore motivated by the incentive to learn more about the lithology and drill less. The principal challenge with building an accurate surface model is that the sedimentary rock mass is coarsely sampled by drilling exploration holes which are typically a long distance apart. Thus, interpolation does not capture adequately local changes in the underlying geology. With the recent advent of consistent and reliable real-time identification of geological boundaries under field conditions using measure-while-drilling data, we pose the local model estimation problem in an adaptive sampling framework. The proposed sampling strategy consists of two phases. First, blast-holes are drilled to the geological boundary of interest, and their locations are adaptively selected to maximize utility in terms of the incremental improvement that can be made to the evolving spatial model. The second phase relies on the predicted geology and drills to an expert based pre-specified standoff distance from the geological boundary of interest, to optimize blasting and minimize its damage. Using data acquired from a coal mine survey bench in Australia, we demonstrate that adaptively choosing blast-holes in Phase 1 can minimize the total number of holes drilled to the top of the coal seam, as opposed to random hole selection, whilst optimizing blasting by maintaining a reasonable compromise in the error in the stopping distances from the seam. We also show that adaptive sampling requires, for accurate estimation, only a fraction of the holes that were initially drilled for this particular dataset.

Outdoor robotic systems rely on perception modules that must be robust to variations in environmental conditions. In particular, vision-based perception systems are affected by illumination variations caused by occlusions. We propose an approach to calculate the lighting distribution of outdoor scenes. The new approach enables us to compensate for shadows and therefore obtain images which are invariant to the sun position and scene geometry, while also retaining the dimensionality of the original data. The method combines images with geometric information provided by range sensors to infer shadows. We select a pair of points on a shadow boundary from a single material and estimate a terrestrial sunlight-skylight ratio. Individual scaling factors are then calculated for all points based on their orientation and incident illumination sources. The result is a coloured point cloud that is independent of illumination variation due to occlusions and geometry. To demonstrate the effectiveness and generalisation of the approach, we present evaluations using two datasets with different cameras. The first uses a hyperspectral sensor that allows us to analyse the results for a large number of wavelengths, while the second dataset uses a standard RGB camera. The approach is shown to consistently provide good illumination compensation in both scenarios.

This paper addresses the problem of multitarget tracking with multi-modal sensing systems for the purpose of robust and accurate localisation of personnel and vehicles. The ability to exploit the redundancy of multi-modal sensing systems for the localisation of a varying number of personnel and vehicles is important for automated environments. In this paper, we introduce the weighted maximum-likelihood extended Kalman filter (WMLEKF) as a multitarget tracker with the aim of introducing a real-time tracking solution with the ability to fuse observations from multi-modal sensors. The WMLEKF introduces a computationallyefficient tracking algorithm that can robustly deal with target initiation and termination while maintaining identity consistency. It relies on a data association method that naturally results in principled track initiation and pruning. It also presents a method of fusion between multi-modal sensors. The WMLEKF was tested on real multi-modal sensor data with results demonstrating its high tracking and track initiation performance, its multi-modal sensor data fusion capability and its ability to maintain identity consistency and correct sensor handover.

In vegetated environments, reliable obstacle detection remains a challenge for state-of-the-art methods, which are usually based on geometrical representations of the environment built from LIDAR and/or visual data. In many cases, in practice field robots could safely traverse through vegetation, thereby avoiding costly detours. However, it is often mistakenly interpreted as an obstacle. Classifying vegetation is insufficient since there might be an obstacle hidden behind or within it. Some Ultra-wide band (UWB) radars can penetrate through vegetation to help distinguish actual obstacles from obstacle-free vegetation. However, these sensors provide noisy and low-accuracy data. Therefore, in this work we address the problem of reliable traversability estimation in vegetation by augmenting LIDAR-based traversability mapping with UWB radar data. A sensor model is learned from experimental data using a support vector machine to convert the radar data into occupancy probabilities. These are then fused with LIDAR-based traversability data. The resulting augmented traversability maps capture the fine resolution of LIDAR-based maps but clear safely traversable foliage from being interpreted as obstacle. We validate the approach experimentally using sensors mounted on two different mobile robots, navigating in two different environments.

• Monitor-while-drilling and geophysical logging data have been acquired for blast holes in an open cut coal mine • Comparison of results show that MWD results respond to the variations in geology revealed by the geophysical logs. • MWD data can be used to predict sonic velocity . • MWD data can be used for geological and geotechnical analysis including blast design.

This paper describes a novel measure called Modulated Specific Energy (SEM) which has been developed for the purpose of characterizing drilled material in open-pit coal mining. In Monitor-While-Drilling (MWD), the information available for coal detection are limited to a small set of drilling parameters that can be measured on a rotary drill rig. Despite this constraint, our analysis shows that MWD can still detect the top of the coal seam consistently without relying on geophysical data — such as bulk density and natural gamma — by using the SEM measure. The proposal utilizes a hypothesized link between a derived drill performance indicator (the rotation-to-thrust power ratio) and geomechanical properties of sedimentary rock strata (shear and compressive strengths) to increase the coal discriminative power of SEM relative to Teale׳s specific energy measure. Its efficacy is demonstrated using mutual information, a simple threshold strategy and an artificial neural network. The results show the SEM can detect the coal seam interface consistently with a greater margin for error, and overcome the problems of low specificity and high variability observed in existing MWD approaches. By reducing the detection uncertainty, the SEM is able to provide consistent feedback while drilling and eliminate trial-and-error. This makes coal mining processes more integrated and reliable, which in turn improves operational effectiveness and efficiency in coal recovery.

Robots have a finite supply of resources such as fuel, battery charge, and storage space. The aim of the Stochastic Collection and Replenishment (SCAR) scenario is to use dedicated agents to refuel, recharge, or otherwise replenish robots in the field to facilitate persistent autonomy. This paper explores the optimisation of the SCAR scenario with a single replenishment agent, using several different objective functions. The problem is framed as a combinatorial optimisation problem, and A* is used to find the optimal schedule. Through a computational study, a ratio objective function is shown to have superior performance compared with a total weighted tardiness objective function, with a greater performance advantage present when using shorter schedule lengths. The importance of incorporating uncertainty in the objective function used in the optimisation process is also highlighted, in particular for scenarios in which the replenishment agent is under- or fully-utilised.

This paper addresses the problem of ensuring mobile robot motion safety when reacting to soft and hard hazards in a static environment. The work is aimed at off-road navigation for mobile ground robots where soft hazards are posed by varying terrain conditions (e.g. deformable soil, slopes, vegetation). Soft hazards pose operating constraints (i.e. speed limits) to the mobile robot that need to be satisfied to ensure motion safety. This paper presents a new morphological erosion operator that generalizes binary obstacle growing to mobility space (the space of speed limits) to deal with both hard and soft hazards seamlessly. This ensures that topological constraints due to vehicle size as well as momentum are taken into account, and leads to a straight-forward approach to generalize the concept of `regions of inevitable collision' for soft hazards.