The maturity of cannabis flowers at harvest critically influences cannabinoid yield and product quality. However, conventional assessment methods rely on subjective visual inspection of trichomes and stigmas, making them inherently inconsistent. This research presents an automated framework integrating computer vision and deep learning to objectively evaluate cannabis flower maturity. High-resolution macro images were acquired using low-cost smartphone-based systems under greenhouse and laboratory conditions. A two-stage pipeline was implemented: a fine-tuned Faster R-CNN model detected trichomes (Precision: 0.815; Recall: 0.802), while a YOLOv8 classifier categorized them into clear, milky, or amber classes (Accuracy: 98.6%). In parallel, a YOLOv8 segmentation model delineated stigmas (AP50: 52.2%) to compute color ratios as maturity indicators. Features were aggregated at the flower level and correlated with HPLC-measured cannabinoid concentrations. A dataset of over 14,000 images was collected across multiple imaging sessions to support training, evaluation, and correlation experiments. Results demonstrated that stigma coloration-detectable with low-end devices-provides a robust visual indicator of peak chemical maturity, with the green-to-orange transition aligning with maximum cannabinoid concentration. This work offers a scalable, cost-effective solution for real-time maturity assessment in cannabis cultivation, contributing to improved harvest timing and quality control.

We consider the problem of solving qualitative decentralized partially observable Markov decision processes (QDec-POMDPs) with deterministic actions. QDec-POMDPs model dynamic systems consisting of a collaborative team of agents acting under uncertainty and partial observability, attempting to reach a desirable goal state. They can be viewed as the multi-agent version of the contingent planning model. In this work, we extend the idea of factored planning from fully observable multi-agent planning to partial observability. Our method operates as follows: First, we simplify the multi-agent planning (MAP) problem by reducing it to a single-agent planning problem. Then, we use the solution to this single-agent problem as a skeleton plan, that each agent attempts to complete separately. We describe different variants of this idea, and in particular, suggest a method that models information about each agent’s knowledge and incorporates the idea of signaling information to other agents through actions. We perform an extensive empirical evaluation over new and old domains, demonstrating the enhanced scalability of our method.

In the multi-agent path finding (MAPF) problem, a group of agents search in a graph for a path for each agent where no two paths collide. While in all applications of MAPF the agents must not collide with each other, in some of them the agents may not wish to share their paths due to privacy constraints. In this work, we formulate two types of privacy constraints for MAPF and propose algorithms that preserve them. The first type of privacy we consider is planning-level privacy, which means that during planning, the agents cannot identify exactly the planned location of the other agents. We propose a general framework for obtaining planning-level privacy, which works by adding mock agents to the planning process. The second type of privacy we consider is execution-level privacy, which is relevant when agents have limited sensing capabilities. Execution-level privacy is preserved if none of the agents is allowed to sense the location of the other agents during execution. We show how to adapt two popular MAPF algorithms, namely PIBT and LaCAM, such that they preserve execution-level privacy. Lastly, we propose a post-processing technique that allows the agents to reduce the sum of costs of the returned solution without losing any privacy. We also implemented our algorithms and evaluated them empirically, showing that the proposed post-processing technique indeed improved cost significantly.

Automated plant phenotyping can help to monitor the growth process of crops, eliminating the high costs associated with traditional manual approaches. Using low-cost devices (e.g., digital cameras), RGB images can be captured under field or greenhouse conditions to track various phenotypes. In this paper, we focused on a particular task - tracking plant growth by identifying and monitoring plant nodes in greenhouse-grown crops. We used a setup where a digital camera captured images at 1-h intervals, with object detection algorithms employed to facilitate rapid and cost-effective tracking of nodes. The main challenge addressed in this paper involved tracking nodes that were hidden temporarily caused by diurnal leaf movements - leaves obscure some nodes at different times throughout the day. Because a node may be hidden for a few hours but visible at other times during the day, one can predict its location while it is hidden. We proposed two approaches, clustering and linear interpolation, for estimating hidden node locations. We collected a set of greenhouse datasets for different crops and conducted empirical comparisons of our methods. Results showed that our approach predicted the node location with an average error of less than 4 cm.

INTRODUCTION:Perturbation-based balance training reduces fall rates dramatically by triggering and improving balance recovery skills. We aimed to investigate whether multidirectional surface perturbation treadmill training, which explicitly challenges age-related impairments in reactive responses, can improve balance recovery responses in standing and reduce annual falls. METHODS:This was a two-arm parallel-group randomized controlled trial with concealed allocation, blinded assessors, data analyzers, and intention-to-treat analysis. Fifty-three older adults aged 80.1 ± 5.2 years, living in retirement housing, were randomized into two groups: (1) surface perturbation-based hands-free treadmill training (SPTT, n = 27) and (2) control group, hands-free treadmill walking training without perturbations (TT, n = 26). Both received a 12-week, 24-session training program. For primary outcomes, we evaluated balance recovery, pre- and post-intervention, by stepping thresholds, percentage of stepping responses, total probability of stepping, and kinematics of reactive stepping to lateral-surface perturbations in standing. Fall incidents were monitored prospectively 1 year after training for a secondary outcome. RESULTS:Both groups showed a significant decrease in the percentage of multiple-step responses (p = 0.013) and a shorter total recovery time to recover balance (p = 0.006). Compared with the TT, the SPTT led to a more significant reduction in single-step and multiple-step thresholds (p = 0.003 and p = 0.002, respectively), total probability of stepping (p = 0.008), shorter first-step length (p = 0.003), total steps path length (p = 0.007), and decreased total center-of-mass (CoM) displacement (p = 0.040) during recovery stepping. One-year prospective fall monitoring revealed nine fall events in the SPTT group compared to 17 in the TT group. Although these numbers are insignificant, they imply a potential generalization that SPTT can reduce annual falls. CONCLUSION:A 12-week SPTT reduces the risk of falls by improving reactive balance responses in retirement-housing older adults. Findings suggest that the primary benefit of SPTT was better control in the CoM following perturbations. This study addressed the generalizability of PBT benefits from walking to standing and the personalization of perturbation training to enhance effectiveness and real-life applicability.

Professional bicycle racing is a popular sport that has attracted significant attention in recent years. The evolution and ubiquitous use of sensors allow cyclists to measure many metrics including power, heart rate, speed, cadence, and more in training and racing. In this paper we explore for the first time assignment of a subset of a team's cyclists to an upcoming race. We introduce RaceFit, a model that recommends, based on recent workouts and past assignments, cyclists for participation in an upcoming race. RaceFit consists of binary classifiers that are trained on pairs of a cyclist and a race, described by their relevant properties (features) such as the cyclist's demographic properties, as well as features extracted from his workout data from recent weeks; as well additional properties of the race, such as its distance, elevation gain, and more. Two main approaches are introduced in recommending on each stage in a race and aggregate from it to the race, or on the entire race. The model training is based on binary label which represent participation of cyclist in a race (or in a stage) in past events. We evaluated RaceFit rigorously on a large dataset of three pro-cycling teams' cyclists and race data achieving up to 80% precision@i. The first experiment had shown that using TP or STRAVA data performs the same. Then the best-performing parameters of the framework are using 5 weeks time window, imputation was effective, and the CatBoost classifier performed best. However, the model with any of the parameters performed always better than the baselines, in which the cyclists are assigned based on their popularity in historical data. Additionally, we present the top-ranked predictive features.

Multi-Agent Pathfinding (MAPF) is the problem of finding paths for multiple agents where each agent aims to reach a given goal location without conflicting with the other agents. In MAPF applications with physical robots, we can expect the agents to have stochastic behavior and imperfect localization. Planning for such centrally-controlled agents can be viewed as a special case of Partially Observable Markov Decision Process (POMDP), but off-the-shelf POMDP solvers cannot scale to plan for even a very small number of agents, due to the exponentially large size of the state and action spaces. Instead, we propose the Online Prioritized Planning (OPP) approach, where each agent computes and follows its individually-optimal policy until a potential conflict is detected. OPP resolves detected potential conflicts by replanning online for a subset of the agents so as to avoid positions that are potentially occupied by other agents. We describe how OPP can be implemented and propose two extensions that encourage the agents to leverage localization actions when needed. We evaluate OPP and its extensions empirically to highlight the pros and cons of our approach and show it can scale better than an offline baseline.

Partially observable Markov decision processes (POMDP) are a useful model for decision-making under partial observability and stochastic actions. Partially Observable Monte-Carlo Planning (POMCP) is an online algorithm for deciding on the next action to perform, using a Monte-Carlo tree search approach, based on the UCT algorithm for fully observable Markov-decision processes. POMCP develops an action-observation tree, and at the leaves, uses a rollout policy to provide a value estimate for the leaf. As such, POMCP is highly dependent on the rollout policy to compute good estimates, and hence identify good actions. Thus, many practitioners who use POMCP are required to create strong, domain-specific heuristics. In this paper, we model POMDPs as stochastic contingent planning problems. This allows us to leverage domain-independent heuristics that were developed in the planning community. We suggest two heuristics, the first is based on the well-known $$h_{add}$$ h add heuristic from classical planning, and the second is computed in belief space, taking the value of information into account.

In multi-agent path finding (MAPF), agents navigate to their target positions without conflict within an environment, typically represented as a graph. Traditionally, the input graph is assumed to be accurate. We investigate MAPF scenarios where the input graph may be inaccurate, containing non-existent edges or missing edges present in the environment. Agents can verify the existence or non-existence of an edge only by moving close to it. To navigate such maps, we propose an online approach where planning and execution are interleaved. As agents gather new information about the environment over time, they replan accordingly. To minimize replanning efforts, we developed methods to identify and replan only for agents affected by observed changes. To scale to larger problems, we defer conflicts resolution expected only in the distant future and adapt single-agent path-finding algorithms to account for map inaccuracies. Experimental results show impressive scalability, solving problems involving over 1000 agents in under 3 minutes.

Acting to complete tasks in stochastic partially observable domains is an important problem in artificial intelligence, and is often formulated as a goal-based POMDP. Goal-based POMDPs can be solved using the RTDP-BEL algorithm, that operates by running forward trajectories from the initial belief to the goal. These trajectories can be guided by a heuristic, and more accurate heuristics can result in significantly faster convergence. In this paper, we develop a heuristic function that leverages the structured representation of domain models. We compute, in a relaxed space, a plan to achieve the goal, while taking into account the value of information, as well as the stochastic effects. We provide experiments showing that while our heuristic is slower to compute, it requires an order of magnitude less trajectories before convergence. Overall, it thus speeds up RTDP-BEL, particularly in problems where significant information gathering is needed.

In multi-agent path finding (MAPF), several agents must move from their current positions to their target positions without colliding. Prior work on MAPF commonly assumed perfect knowledge of the environment. We consider a MAPF setting where this is not the case, and the planner does not know a-priori whether some positions are blocked or not. To sense whether such a position is traversable, an agent must move close to it and adapt its behavior accordingly. In this work we focus on solving this type of MAPF problem, for cases where planning is centralized but cannot be done during execution. In this setting, a solution can be formulated as a plan tree for each agent, branching on the observations. We propose algorithms for finding such plans trees for two modes of executions: centralized, where the agents share information concerning observed obstacles during execution, a decentralized, where such communication is not allowed. The proposed algorithms are complete and can be configured to optimize solution cost, measured for either the best case or the worst case. We implemented these algorithms and provide experimental results demonstrating how our approach scales with respect to the number of agents and the number of positions we are uncertain about. The results show that our algorithms can solve non-trivial problems, but also highlight that this type of MAPF problems is significantly harder than classical MAPF.

In citizen science, regular people provide invaluable information by contributing to scientific projects. Citizen science platforms, such as SciStarter, provide easy access to numerous such projects. Often, users contribute mainly to a relatively small set of popular projects, while it is difficult for many projects to draw the attention of users. Thus, increasing the contribution of users to such low-popularity projects may increase scientific and societal impact. In this paper, we explore the power of a recommender system to draw attention to less popular projects. Standard use of recommendation systems often leads to limited exposure of less popular (tail) projects. We thus propose a re-ranking approach based on “lift boosting,” which uses the statistical lift measure to enhance the exposure of tail projects. By combining lift and traditional relevance measures, our method re-ranks the recommendation list to emphasize projects that are both relevant to the user while also have a high lift value. We implement our approach on SciStarter, one of the biggest citizen science platforms on the web. We conduct an online experiment involving over 2000 real users. Our results show a positive shift towards less popular projects without compromising overall contribution rates. This work demonstrates the potential of our lift-boosting method for promoting the discovery of tail projects in citizen science platforms, thereby fostering a more diverse range of scientific contributions.

Traditionally, a contingent plan, branching on the observations an agent obtains throughout plan execution, must reach a goal state from every possible initial state. However, in many real world problems, no such plan exists. Yet, there are plans that reach the goal from some initial states only. From the other initial states, they eventually reach a deadend—a state from which the goal can not be achieved. Deadends that cannot be avoided by resorting to a different plan, are called unavoidable deadends. In this paper we study planning with unavoidable deadends in belief space. We distinguish between two types of such deadends, and adapt offline and online contingent planners to identify and handle unavoidable deadends, using two approaches—an active approach that begins by distinguishing between the solvable and deadend states, and a lazy approach, that plans to achieve the goal, identifying deadends as they occur. We empirically analyze how each approach performs in different cases.

Mobile health aims to enhance health outcomes by delivering interventions to individuals as they go about their daily life. The involvement of care partners and social support networks often proves crucial in helping individuals managing burdensome medical conditions. This presents opportunities in mobile health to design interventions that target the dyadic relationship -- the relationship between a target person and their care partner -- with the aim of enhancing social support. In this paper, we develop dyadic RL, an online reinforcement learning algorithm designed to personalize intervention delivery based on contextual factors and past responses of a target person and their care partner. Here, multiple sets of interventions impact the dyad across multiple time intervals. The developed dyadic RL is Bayesian and hierarchical. We formally introduce the problem setup, develop dyadic RL and establish a regret bound. We demonstrate dyadic RL's empirical performance through simulation studies on both toy scenarios and on a realistic test bed constructed from data collected in a mobile health study.

Multi-agent collaboration under partial observability is a difficult task. Multi-agent reinforcement learning (MARL) algorithms that do not leverage a model of the environment struggle with tasks that require sequences of collaborative actions, while Dec-POMDP algorithms that use such models to compute near-optimal policies, scale poorly. In this paper, we suggest the Team-Imitate-Synchronize (TIS) approach, a heuristic, model-based method for solving such problems. Our approach begins by solving the joint team problem, assuming that observations are shared. Then, for each agent we solve a single agent problem designed to imitate its behavior within the team plan. Finally, we adjust the single agent policies for better synchronization. Our experiments demonstrate that our method provides comparable solutions to Dec-POMDP solvers over small problems, while scaling to much larger problems, and provides collaborative plans that MARL algorithms are unable to identify.

Tumor development is clinically characterized through the manual review of histopathological Whole Slide Images (WSI). However, the molecular attributes influencing tumor morphology are not entirely comprehended. Here, we present RNALerner, an innovative tool designed to expedite the identification of correlations between gene expression and tumor morphology as presented in H&E WSI. RNALerner achieves its efficiency by transforming the problem from linear regression to binary classification of high versus low RNA levels, and the use of Resnet18, Convolutional Neural Network (CNN) model. Furthermore, the training phase of the model is halted after only 3 iterations. Upon comparing our results with previous work, we discovered a similar number of statistically significant correlated genes but with a reduction in the number of model parameters and processing time. Analysis of the significant pathways revealed both similarities to and deviations from earlier findings, bringing forth new pathways in the process. RNALerner represents an advancement toward the practical integration of machine learning in WSI analysis, which holds the potential to substantially improve disease diagnosis and guide more effective treatments.

Partially observable Markov decision processes (POMDP) are a useful model for decision-making under partial observability and stochastic actions. Partially Observable Monte-Carlo Planning is an online algorithm for deciding on the next action to perform, using a Monte-Carlo tree search approach, based on the UCT (UCB applied to trees) algorithm for fully observable Markov-decision processes. POMCP develops an action-observation tree, and at the leaves, uses a rollout policy to provide a value estimate for the leaf. As such, POMCP is highly dependent on the rollout policy to compute good estimates, and hence identify good actions. Thus, many practitioners who use POMCP are required to create strong, domain-specific heuristics. In this paper, we model POMDPs as stochastic contingent planning problems. This allows us to leverage domain-independent heuristics that were developed in the planning community. We suggest two heuristics, the first is based on the well-known h_add heuristic from classical planning, and the second is computed in belief space, taking the value of information into account.

Introduction The inability to recover from unexpected lateral loss of balance may be particularly relevant to the problem of falling. Aim We aimed to explore whether different kinematic patterns and strategies occur in the first recovery step in single-step trials in which a single step was required to recover from a fall, and in multiple-step trials in which more than one step was required to recover from a fall. In addition, in the multiple-step trials, we examined kinematic patterns of balance recovery where extra steps were needed to recover balance. Methods Eighty-four older adults (79.3 ± 5.2 years) were exposed to unannounced right/left perturbations in standing that were gradually increased to trigger a recovery stepping response. We performed a kinematic analysis of the first recovery step of all single-step and multiple-step trials for each participant and of total balance recovery in the multiple-step trial. Results Kinematic patterns and strategies of the first recovery step in the single-step trials were significantly dependent on the perturbation magnitude. It took a small, yet significantly longer time to initiate a recovery step and a significantly longer time to complete the recovery step as the magnitude increased. However, the first recovery step in the multiple-step trials showed no significant differences between different perturbation magnitudes; while, in total balance recovery of these trials, we observed a small, yet significant difference as the magnitude increased. Conclusions At relatively low perturbation magnitudes, i.e., single-step trials, older adults selected different first stepping strategies and kinematics as perturbation magnitudes increased, suggesting that this population activated pre-planned programs based on the perturbation magnitude. However, in the first recovery step of the multiple-step trials, i.e., high perturbation magnitudes, similar kinematic movement patterns were used at different magnitudes, suggesting a more rigid, automatic behavior, while the extra-steps were scaled to the perturbation magnitude. This suggest that older adults activate pre-planned programs based on the magnitude of the perturbation, even before the first step is completed..

Collaborative privacy preserving planning ( cppp ) gained much attention in the past decade. cppp aims to create solutions for multi agent planning problems where cooperation is required to achieve an efficient solution, without exposing information that the agent considers private in the process. To date, cppp has focused on domains with deterministic action effects. However, in real-world problems action effects are often non-deterministic, and actions can have multiple possible effects with varying probabilities. In this paper, we introduce Stochastic cppp ( scppp ), which is an extension of cppp to domains with stochastic action effects. We show how scppp can be modeled as a Markov decision process ( mdp ) and how the value-iteration algorithm can be adapted to solve it. This adaptation requires extending value-iteration to support multiple agents and privacy. Then, we present two adaptions of the real-time dynamic programming ( rtdp ) algorithm, a popular algorithm for solving mdp s, designed to solve scppp problems. The first rtdp adaptation, called distributed rtdp ( drtdp ), yields identical behavior to applying rtdp in a centralized manner on the joint problem. To preserve privacy, drtdp uses a message passing mechanism adopted from the mafs algorithm. The second rtdp adaptation is an approximation of drtdp called public synchronization rtdp ( ps - rtdp ). ps - rtdp differs from drtdp mainly in its message passing mechanism, where ps - rtdp sends significantly fewer messages than drtdp . We experimented on domains adapted from the deterministic cppp literature by adding different stochastic effects to different actions. The results show that ps - rtdp can reduce the amount of messages compared to drtdp by orders of magnitude thus improving run-time, while producing policies with similar expected costs.

Citizen science projects rely on volunteers to contribute time and effort to solve scientific problems, but the majority of citizen science users typically contribute to only one or two projects. Recommender systems have been recently used in citizen science to motivate people to contribute to additional projects. However, these systems often recommend the more popular projects at the expense of less popular projects which need people's contributions more critically. In this work we develop a post processing approach for enhancing "long-tail" item recommendation that can be applied to any recommendation system. We propose a novel re-ranking model, based on the lift measure used in machine learning, which considers item co-occurrence as well as popularity for enhancing long tail recommendations. We demonstrate the efficacy of our approach in the citizen science domain on two data sets and three state of the art recommendation algorithms, comparing hit rate before and after applying lift boosting. Additionally, we compare our approach to two predetermined re-ranking baselines. Results show that our proposed approach significantly improves performance for tail item recommendation without a substantial loss in head item and overall item recommendation performance. Our approach is general and can be naturally applied to existing recommendation systems in citizen science that personalize project suggestions to users, potentially leading to an increase in efficiency and performance.