Partially Observable Markov Decision Processes (POMDPs) are a powerful framework for planning under uncertainty. They allow to model state uncertainty as a belief probability distribution. Approximate solvers based on Monte Carlo sampling show great success to relax the computational demand and perform online planning. However, scaling to complex realistic domains with many actions and long planning horizons is still a major challenge, and a key point to achieve good performance is guiding the action-selection process with domain-dependent policy heuristics which are tailored for the specific application domain. We propose to learn high-quality heuristics from POMDP traces of executions generated by any solver. We convert the belief-action pairs to a logical semantics, and exploit data- and time-efficient Inductive Logic Programming (ILP) to generate interpretable belief-based policy specifications, which are then used as online heuristics. We evaluate thoroughly our methodology on two notoriously challenging POMDP problems, involving large action spaces and long planning horizons, namely, rocksample and pocman. Considering different state-of-the-art online POMDP solvers, including POMCP, DESPOT and AdaOPS, we show that learned heuristics expressed in Answer Set Programming (ASP) yield performance superior to neural networks and similar to optimal handcrafted task-specific heuristics within lower computational time. Moreover, they well generalize to more challenging scenarios not experienced in the training phase (e.g., increasing rocks and grid size in rocksample, incrementing the size of the map and the aggressivity of ghosts in pocman).

Offline Reinforcement Learning (RL) allows policies to be trained on pre-collected datasets without requiring additional interactions with the environment. This approach bypasses the need for real-time data acquisition in real-world applications, which can be impractical due to the safety issues inherent in the learning process. However, offline RL faces significant challenges, such as distributional shifts and extrapolation errors, and the resulting policies might underperform compared to the baseline policy. Safe policy improvement algorithms mitigate these issues, enabling the reliable deployment of RL approaches in real-world scenarios where historical data is available, guaranteeing that any policy changes will not result in worse performance compared to the baseline policy used to collect training data. In this paper, we propose MCTS-SPIBB, an algorithm that leverages Monte Carlo Tree Search (MCTS) for scaling safe policy improvement to large domains. We theoretically prove that the policy generated by MCTS-SPIBB converges to the optimal safely improved policy produced by Safe Policy Improvement with Baseline Bootstrapping (SPIBB) as the number of simulations increases. Additionally, we introduce SDP-SPIBB, a novel extension of SPIBB designed to address the scalability limitations of the standard algorithm via Scalable Dynamic Programming. Our empirical analysis across four benchmark domains demonstrates that MCTS-SPIBB and SDP-SPIBB significantly enhance the scalability of safe policy improvement, providing robust and efficient algorithms for large-scale applications. These contributions represent a significant step towards the deployment of safe RL algorithms in complex real-world environments.

In this work, we focus on safe policy improvement in multi-agent domains where current state-of-the-art methods cannot be effectively applied because of large state and action spaces. We consider recent results using Monte Carlo Tree Search for Safe Policy Improvement with Baseline Bootstrapping and propose a novel algorithm that scales this approach to multi-agent domains, exploiting the factorization of the transition model and value function. Given a centralized behavior policy and a dataset of trajectories, our algorithm generates an improved policy by selecting joint actions using a novel extension of Max-Plus (or Variable Elimination) that constrains local actions to guarantee safety criteria. An empirical evaluation on multi-agent SysAdmin and multi-UAV Delivery shows that the approach scales to very large domains where state-of-the-art methods cannot work.

Vision-and-Language Navigation in Continuous Environments (VLN-CE) is one of the most intuitive yet challenging embodied AI tasks. Agents are tasked to navigate towards a target goal by executing a set of low-level actions, following a series of natural language instructions. All VLN-CE methods in the literature assume that language instructions are exact. However, in practice, instructions given by humans can contain errors when describing a spatial environment due to inaccurate memory or confusion. Current VLN-CE benchmarks do not address this scenario, making the state-of-the-art methods in VLN-CE fragile in the presence of erroneous instructions from human users. For the first time, we propose a novel benchmark dataset that introduces various types of instruction errors considering potential human causes. This benchmark provides valuable insight into the robustness of VLN systems in continuous environments. We observe a noticeable performance drop (up to −25%) in Success Rate when evaluating the state-of-the-art VLN-CE methods on our benchmark. Moreover, we formally define the task of Instruction Error Detection and Localization, and establish an evaluation protocol on top of our benchmark dataset. We also propose an effective method, based on a cross-modal transformer architecture, that achieves the best performance in error detection and localization, compared to baselines. Surprisingly, our proposed method has revealed errors in the validation set of the two commonly used datasets for VLN-CE, i.e., R2R-CE and RxR-CE, demonstrating the utility of our technique in other tasks.

In the Vision-and-Language Navigation in Continuous Environments (VLN-CE) task, the human user guides an autonomous agent to reach a target goal via a series of low-level actions following a textual instruction in natural language. However, most existing methods do not address the likely case where users may make mistakes when providing such instruction (e.g., "turn left" instead of "turn right"). In this work, we address a novel task of Interactive VLN in Continuous Environments (IVLN-CE), which allows the agent to interact with the user during the VLN-CE navigation to verify any doubts regarding the instruction errors. We propose an Interactive Instruction Error Detector and Localizer (I2EDL) that triggers the user-agent interaction upon the detection of instruction errors during the navigation. We leverage a pre-trained module to detect instruction errors and pinpoint them in the instruction by cross-referencing the textual input and past observations. In such way, the agent is able to query the user for a timely correction, without demanding the user's cognitive load, as we locate the probable errors to a precise part of the instruction. We evaluate the proposed I2EDL on a dataset of instructions containing errors, and further devise a novel metric, the Success weighted by Interaction Number (SIN), to reflect both the navigation performance and the interaction effectiveness. We show how the proposed method can ask focused requests for corrections to the user, which in turn increases the navigation success, while minimizing the interactions.

Partially Observable Markov Decision Processes (POMDPs) are a powerful framework for planning under uncertainty. They allow to model state uncertainty as a belief probability distribution. Approximate solvers based on Monte Carlo sampling show great success to relax the computational demand and perform online planning. However, scaling to complex realistic domains with many actions and long planning horizons is still a major challenge, and a key point to achieve good performance is guiding the action-selection process with domain-dependent policy heuristics which are tailored for the specific application domain. We propose to learn high-quality heuristics from POMDP traces of executions generated by any solver. We convert the belief-action pairs to a logical semantics, and exploit data- and time-efficient Inductive Logic Programming (ILP) to generate interpretable belief-based policy specifications, which are then used as online heuristics. We evaluate thoroughly our methodology on two notoriously challenging POMDP problems, involving large action spaces and long planning horizons, namely, rocksample and pocman. Considering different state-of-the-art online POMDP solvers, including POMCP, DESPOT and AdaOPS, we show that learned heuristics expressed in Answer Set Programming (ASP) yield performance superior to neural networks and similar to optimal handcrafted task-specific heuristics within lower computational time. Moreover, they well generalize to more challenging scenarios not experienced in the training phase (e.g., increasing rocks and grid size in rocksample, incrementing the size of the map and the aggressivity of ghosts in pocman).

We propose a solution for Active Visual Search of objects in an environment, whose 2D floor map is the only known information. Our solution has three key features that make it more plausible and robust to detector failures compared to state-of-the-art methods: i) it is unsupervised as it does not need any training sessions. ii) During the exploration, a probability distribution on the 2D floor map is updated according to an intuitive mechanism, while an improved belief update increases the effectiveness of the agent's exploration. iii) We incorporate the awareness that an object detector may fail into the aforementioned probability modelling by exploiting the success statistics of a specific detector. Our solution is dubbed POMP-BE-PD (Pomcp-based Online Motion Planning with Belief by Exploration and Probabilistic Detection). It uses the current pose of an agent and an RGB-D observation to learn an optimal search policy, exploiting a POMDP solved by a Monte-Carlo planning approach. On the Active Vision Dataset Benchmark, we increase the average success rate over all the environments by a significant 35 % while decreasing the average path length by 4 % with respect to competing methods. Thus, our results are state-of-the-art, even without any training procedure.

COVID-19 had substantial effects on the IoT community which designs systems for safety: the urge to face masks worn by everyone, the analysis of crowds to avoid the spread of the disease, and the sanitization of public environments has led to exceptional research acceleration and fast engineering of the related solutions. Now that the pandemic is losing power, some applications are becoming less important, while others are proving to be useful regardless of the criticality of COVID-19. The Safe Place project is a prime example of this situation (DATE23 MPP category: final stage). Safe Place is an Italian 3M euro regional industrial/academic project, financed by European funds, created to ensure a multidisciplinary choral reaction to COVID-19 in critical environments such as rest homes and public places. Safe Place consortium was able to understand what is no longer useful in this post-pandemic period, and what instead is potentially attractive for the market. For example, the detection of face masks has little importance, while sanitization does have much. This paper shares such analysis, which emerged through a co-design process of three public Safe Place project demonstrators, involving heterogeneous figures spanning from scientists to lawyers.

Algorithms for safely improving policies are important to deploy reinforcement learning approaches in real-world scenarios. In this work, we propose an algorithm, called MCTS-SPIBB, that computes safe policy improvement online using a Monte Carlo Tree Search based strategy. We theoretically prove that the policy generated by MCTS-SPIBB converges, as the number of simulations grows, to the optimal safely improved policy generated by Safe Policy Improvement with Baseline Bootstrapping (SPIBB), a popular algorithm based on policy iteration. Moreover, our empirical analysis performed on three standard benchmark domains shows that MCTS-SPIBB scales to significantly larger problems than SPIBB because it computes the policy online and locally, i.e., only in the states actually visited by the agent.

Partially Observable Monte Carlo Planning (POMCP) is a powerful online algorithm that can generate approximate policies for large Partially Observable Markov Decision Processes. The online nature of this method supports scalability by avoiding complete policy representation. However, the lack of an explicit policy representation hinders interpretability and a proper evaluation of the risks an agent may incur. In this work, we propose a methodology based on Maximum Satisfiability Modulo Theory (MAX-SMT) for analyzing POMCP policies by inspecting their traces, namely, sequences of belief-action pairs generated by the algorithm. The proposed method explores local properties of the policy to build a compact and informative summary of the policy behaviour. Moreover, we introduce a rich and formal language that a domain expert can use to describe the expected behaviour of a policy. In more detail, we present a formulation that directly computes the risk involved in taking actions by considering the high-level elements specified by the expert. The final formula can identify risky decisions taken by POMCP that violate the expert indications. We show that this identification process can be used offline (to improve the policy's explainability and identify anomalous behaviours) or online (to shield the risky decisions of the POMCP algorithm). We present an extended evaluation of our approach on four domains: the well-known tiger and rocksample benchmarks, a problem of velocity regulation in mobile robots, and a problem of battery management in mobile robots. We test the methodology against a state-of-the-art anomaly detection algorithm to show that our approach can be used to identify anomalous behaviours in faulty POMCP. We also show, comparing the performance of shielded and unshielded POMCP, that the shielding mechanism can improve the system's performance. We provide an open-source implementation of the proposed methodologies at https://github.com/GiuMaz/XPOMCP.

In this work, we concentrate on the detection of anomalous behaviors in systems operating in the physical world and for which it is usually not possible to have a complete set of all possible anomalies in advance. We present a data augmentation and retraining approach based on adversarial learning for improving anomaly detection. In particular, we first define a method for generating adversarial examples for anomaly detectors based on Hidden Markov Models (HMMs). Then, we present a data augmentation and retraining technique that uses these adversarial examples to improve anomaly detection performance. Finally, we evaluate our adversarial data augmentation and retraining approach on four datasets showing that it achieves a statistically significant performance improvement and enhances the robustness to adversarial attacks. Key differences from the state-of-the-art on adversarial data augmentation are the focus on multivariate time series (as opposed to images), the context of one-class classification (in contrast to standard multi-class classification), and the use of HMMs (in contrast to neural networks).

In this paper, we focus on the problem of learning online an optimal policy for Active Visual Search (AVS) of objects in unknown indoor environments. We propose POMP++, a planning strategy that introduces a novel formulation on top of the classic Partially Observable Monte Carlo Planning (POMCP) framework, to allow training-free online policy learning in unknown environments. We present a new belief reinvigoration strategy that enables the use of POMCP with a dynamically growing state space to address the online generation of the floor map. We evaluate our method on two public benchmark datasets, AVD that is acquired by real robotic platforms and Habitat ObjectNav that is rendered from real 3D scene scans, achieving the best success rate with an improvement of >10% over the state-of-the-art methods.

Online planning methods for partially observable Markov decision processes (POMDPs) have recently gained much interest. In this paper, we propose the introduction of prior knowledge in the form of (probabilistic) relationships among discrete state-variables, for online planning based on the well-known POMCP algorithm. In particular, we propose the use of hard constraint networks and probabilistic Markov random fields to formalize state-variable constraints and we extend the POMCP algorithm to take advantage of these constraints. Results on a case study based on Rocksample show that the usage of this knowledge provides significant improvements to the performance of the algorithm. The extent of this improvement depends on the amount of knowledge encoded in the constraints and reaches the 50% of the average discounted return in the most favorable cases that we analyzed.

In this paper, we propose the use of subspace clustering to detect the states of dynamical systems from sequences of observations. In particular, we generate sparse and interpretable models that relate the states of aquatic drones involved in autonomous water monitoring to the properties (e.g., statistical distribution) of data collected by drone sensors. The subspace clustering algorithm used is called SubCMedians. A quantitative experimental analysis is performed to investigate the connections between i) learning parameters and performance, ii) noise in the data and performance. The clustering obtained with this analysis outperforms those generated by previous approaches.

An immunological data-driven model is proposed, for age related changes in the network of relationships among cell quantities of eight peripheral B lymphocyte subpopulations, that is, cells exhibiting all combinations of three specific receptor clusters (CD27, CD23, CD5). The model is based on immunological data (quantities of cells exhibiting CD19, characterizing B lymphocytes) from about six thousands patients, having an age ranging between one day and ninety-five years, by means of a suitably combination of data analysis methods, such as piecewise linear regression models. With relaxed values for statistically significant models (coefficient p-values bounded by 0.05), we found a network holding for all ages, that likely represents the general assessment of adaptive immune system for healthy human beings. When statistical validation comes to be more restrictive, we found that some of these interactions are lost with aging, as widely observed in medical literature. Namely, interesting (inverse or directed) proportions are highlighted among mutual quantities of a partition of peripheral B lymphocytes.

MP-GeneticSynth is a Java tool for discovering the logic and regulation mechanisms responsible for observed biological dynamics in terms of finite difference recurrent equations. The software makes use of: (i) metabolic P systems as a modeling framework, (ii) an evolutionary approach to discover flux regulation functions as linear combinations of given primitive functions, (iii) a suitable reformulation of the least squares method to estimate function parameters considering simultaneously all the reactions involved in complex dynamics. The tool is available as a plugin for the virtual laboratory MetaPlab. It has graphical and interactive interfaces for data preparation, a priori knowledge integration, and flux regulator analysis. Availability and implementation: Source code, binaries, documentation (including quick start guide and videos) and case studies are freely available at http://mplab.sci.univr.it/plugins/mpgs/index.html.

Metabolic P systems, also called MP systems, are discrete dynamical systems which proved to be effective for modeling biological systems. Their dynamics is generated by means of a metabolic algorithm based on "flux regulation functions". A significant problem related to the generation of MP models from experimental data concerns the synthesis of these functions. In this paper we introduce a new approach to the synthesis of MP fluxes relying on neural networks as universal function approximators, and on evolutionary algorithms as learning techniques. This methodology is successfully tested in the case study of mitotic oscillator in early amphibian embryos.