In precision agriculture, overhead imagery can help pinpoint locations that appear to be under water stress, in which case sampling leaves or soil in those areas should be prioritized. We formalize the notion of tasks with distinct urgency by assigning a priority level to tasks. Besides the task priority level, deciding a next task for a robot to complete is also dependent on available budget, which can be of multiple types. For instance, the number of locations that a robot can visit and sample from in one ‘trip’ is constrained by both the energy capacity to move between locations and the robot’s sample payload capacity.Exceeding the energy budget can prevent the robot from returning to the base station to recharge and drop collected samples, whereas exceeding the sample payload capacity may cause potential robot and sample damage. Here we consider an energy budget for the robot moving between locations, and a resource budget linked to task execution. The two budgets are independent of each other, and both can be reset to their initial values when the robot returns to the base station. The actual amount of resources consumed to execute a task can differ from what is the expectation in practice. In fact, the actual amount of resources consumed for task execution is revealed only after the task has been completed. To model this,plastic pots large we consider the cost to complete a task to be a stochastic random variable that follows some known distribution. The cost to move between locations, however, is considered to be deterministic. Specific details are given in the following.
This paper introduces a new stochastic task allocation approach, termed Next-Best-Action Planning , for task planning under uncertainty in precision agriculture. The paper also contributes a new Stochastic-Vertex-Cost Aisle Graph . SAG is an extension of the aisle graph, which is often used to describe agriculture-related environments.Using SAG, our proposed NBA-P algorithm simultaneously determines 1) how to optimally schedule which tasks to perform at run-time, and 2) when to optimally stop performing new tasks and return back to the base station also at run-time. Further, it can be extended to multi-robot teams. We test in single- and multi-robot cases using both simulated data and 10 real-world datasets collected in a commercial vineyard at central California. In all cases, NBAP achieves higher efficiency than naive lawnmower, informed lawnmower, and series Greedy Partial Row planners in terms of more return per visited vertices, less resources wasted because of aborted tasks, and less total visited vertices. Aisle graphsmodel motion constraints emerging when robots navigate in structured environments like agricultural fields. Vertices denote task locations, and edges represent connections between locations. Any two rows connect to each other only via the two end vertices. Moving backwards is not allowed; if a robot enters a row, it will have to reach the row’s end before moving to another row. In the original aisle graph, vertices and edges are associated with known and constant reward and movement costs, respectively. Our extension, SAG, can also represent uncertain task costs. Orienteering can tackle persistent sampling on aisle graphs. Orienteering is NP-hard, and thus greedy heuristics are often employed. Recent efforts on stochastic orienteering associate stochastic costs to graph edges and propose a time-aware policy for a robot to adjust its path to avoid exceeding a certain budget. However, addressing cases that involve uncertain task cost on vertices for aisle graphs remains open. NBA-P tackles the problem by simultaneously considering uncertain task costs on vertices and deterministic costs on edges.
Optimal stopping can be used to find the criteria to terminate a process while incorporating uncertainty. Often, data arrive in sequence, and irrevocable decision has to be made as to when the expected return is maximized. Optimal stopping has been used in robotics applications like target tracking and marine ecosystem monitoring. However, no motion constraints, like those imposed by aisle graphs, apply to robot actions, and hence existing methods cannot be ported over to operations on aisle graphs. Paths planned with NBA-P fill the gap, as they directly apply to environments with motion constraints captured by aisle graphs. Our method applies when: 1) motion constraints in the environment can be captured by a SAG; 2) the cost of completing tasks follow exponential distributions; and 3) the obtained gain by completing a task is proportional to the actual task cost. Anthropogenic landscapes and their associated biodiversity are novel ecosystems that are expanding globally. Urbanization and agriculture, particularly, are the products of growing human populations and have accelerated rates of land conversion . These landscapes are typically associated with transformations in habitat structure, climate, and connectivity as well as the establishment of non-native species . However, even with their increasing dominance as a land use, many questions remain regarding the impact of these anthropogenic landscapes on ecological communities. Particularly, although there are differences between these ecosystems and their less human-modified equivalents, there is evidence for the potential of these anthropogenic landscapes to support biodiversity in unexpected ways that in some cases may be equivalent, or even surpass the biodiversity in surrounding natural landscapes . With the continued expansion of human-modified ecosystems coupled with the inability of reserves to support more than a fraction of the world’s biodiversity , understanding the dynamics of communities in these landscapes is necessary to evaluate their conservation potential and opportunities for restoration and management . Specifically, the temporal dynamics, or phenologies, of species have been observed to be vulnerable to disruption .
Changes in the temporal activity of species can lead to mismatches in the presence of species and their biotic or abiotic resources, which can impact the functioning of ecosystems . Like global climate change which has been found to result in phenological shifts , the urban heat island effect is a well-documented local phenomenon experienced as significantly warmer temperatures in cities relative to the surrounding landscape due to higher energy use and impervious surface area . As a result, plants bloom earlier and more densely and bird migration advances earlier in urban contexts .Human-altered landscapes may also shift the activity of wildlife by altering the variety and timing of resource availability. Many natural areas in temperate environments experience a large burst of diverse plant growth in the spring,planting pot seedlings and by the end of the summer, there are very few resources available . Urban areas, while likely having less vegetative cover than less human-modified landscapes, often support many exotic plants, which are supplemented with water and nutrient inputs, allowing for an extended vegetative and flowering season. As a result, urban areas may support more limited but constant vegetative resources throughout the year . In contrast, agricultural landscapes, due to the phenology of monoculture crops, have large patches of dense vegetative resources that fluctuate greatly from early spring to the end of the summer . Such within-year differences in vegetative availability between land-use types have been documented through the use of remote sensing and may affect the seasonal population dynamics of the animal communities that rely on floral resources . Focusing on wild bees, organisms that are highly dependent on floral resources, we ask whether there are differences in the seasonal population dynamics of bee communities in urban, agricultural, and natural landscapes. Although bee seasonality and movement has been documented in urban and agricultural landscapes , differences in the phenological dynamics of bee communities between urban, agricultural, and natural land-use types have not been explored. Here, we evaluate the hypothesis that neighboring land-use types exhibit different patterns in bee community phenology throughout the year. We predict the temporal dynamics of bee communities in urban landscapes will be less variable than in agriculture and less human-modified areas because resources are more stable. We also predict that because the pulse of resources availability in less human-modified and agricultural area occurs at different times, the bee community phenology will shift in the different land uses to track that availability. Understanding how the novel communities in human-modified landscapes function will help to better inform restoration and conservation efforts.Our study landscape was located in east Contra Costa County around Brentwood, California, where natural, agricultural, and urban areas intersect with each other within a 20 9 20 km region . Large areas of land remain protected from development within this region by regional and state parks as well as the local water district’s watershed. This undeveloped land consists mainly of grasslands and oak woodlands, some portions of which are managed for grazing.
East Contra Costa County has had a farming community presence since the late 19th century. The agricultural areas of Brentwood, Knightsen, and Byron mostly consist of orchards , corn, alfalfa, and tomatoes . A housing boom in the 1990s led to massive residential growth in the area. The city of Brentwood has grown from <2500 people in the 1970s to over 50 000 today , and nearby Antioch has over 100 000 residents . Using NOAA’s 2006 Pacific Coast Land Cover data set , a 500-m buffer was created around each site, and the number of pixels classified as agricultural, urban, natural, water, or bare land was extracted. We classified each site based on the dominant land-use type within its 500 m buffer. In 2010, we had 18 sites, with six each classified as types ‘urban’, ‘agricultural’, and ‘natural.’ In 2011 and 2012, we increased to have a total of 24 sites, with eight of each land use classification. Sites were selected to be at least one km away from all others, based on assumed maximum bee foraging ranges . Although certain bee species have been recorded foraging over a kilometer , most bees have nesting and foraging habitat within a few hundred meters of each other . At each site, we designed a standardized pan-trapping transect of 15 bowls spaced five meters apart in alternating colors of fluorescent blue, white, and fluorescent yellow following a modified version of established protocols for pan-trapping bees . Bowls were filled to the brim with soapy water . In 2010, transects were set up during peak bee flying hours for the four-hour period between 10:30 and 14:30 , with four sites sampled per day, and all sites sampled on consecutive days. These 2010 transects were sampled at two collecting periods: once in the early summer and once in the late summer. In 2011 and 2012, transects were left out for a 24-h period , so that more sites could be sampled simultaneously and therefore, 24 sites could be sampled during each collecting period. All 24 sites were sampled within 4 days of each other, during four collecting periods: early spring, late spring, early summer, and late summer. There were a total of 228 collecting events . Long-term sampling methods such as this have not been found to affect bee community structure . The goal of collection was to sample the bee community that was flying through the landscape searching for resources. For this reason, the human-altered sites were deliberately selected so as not to be adjacent to any mass-flowering plants of agricultural crops or gardens to reduce potential pan-trapping biases of bees actively foraging on immediately local resources . All sites were selected in easily accessible, open areas that received full sun. Natural areas were in grassland habitat, so we selected agricultural sites that were either weedy field margin edges or fallow fields, and urban sites that were vacant lots or green ways. The weedy flower margins in urban and agricultural landscapes generally had equivalent flowering levels as those in the natural areas, making the adjacent floral resources similar, allowing the floral availability on a landscape scale to be the primary differentiating factor. Bee specimens were identified to species . The only exception was for bees of the genus Lasioglossum, which due to their overwhelming abundance combined with difficulty of identification, were identified to the genus level. The vast majority of Lasioglossum collected were primitively social generalist species of the sub-genera Dialictus and Evylaeus. Voucher specimens are deposited at the Essig Museum of Entomology at the University of California, Berkeley.