In planning processes of computational decision-making agents, generative or predictive models are often used as "generators" to propose "targets" representing sets of expected or desirable states. Unfortunately, learned models inevitably hallucinate infeasible targets that can cause delusional behaviors and safety concerns. We first investigate the kinds of infeasible targets that generators can hallucinate. Then, we devise a strategy to identify and reject infeasible targets by learning a target feasibility evaluator. To ensure that the evaluator is robust and non-delusional, we adopted a design choice combining off-policy compatible learning rule, distributional architecture, and data augmentation based on hindsight relabeling. Attaching to a planning agent, the designed evaluator learns by observing the agent's interactions with the environment and the targets produced by its generator, without the need to change the agent or its generator. Our controlled experiments show significant reductions in delusional behaviors and performance improvements for various kinds of existing agents.

Multi-View Reinforcement Learning (MVRL) seeks to provide agents with multi-view observations, enabling them to perceive environment with greater effectiveness and precision. Recent advancements in MVRL focus on extracting latent representations from multiview observations and leveraging them in control tasks. However, it is not straightforward to learn compact and task-relevant representations, particularly in the presence of redundancy, distracting information, or missing views. In this paper, we propose Multi-view Fusion State for Control (MFSC), firstly incorporating bisimulation metric learning into MVRL to learn task-relevant representations. Furthermore, we propose a multiview-based mask and latent reconstruction auxiliary task that exploits shared information across views and improves MFSC’s robustness in missing views by introducing a mask token. Extensive experimental results demonstrate that our method outperforms existing approaches in MVRL tasks. Even in more realistic scenarios with interference or missing views, MFSC consistently maintains high performance. The project code is available at https://github.com/zpwdev/MFSC.

Inspired by human conscious planning, we propose Skipper, a model-based reinforcement learning framework utilizing spatio-temporal abstractions to generalize better in novel situations. It automatically decomposes the given task into smaller, more manageable subtasks, and thus enables sparse decision-making and focused computation on the relevant parts of the environment. The decomposition relies on the extraction of an abstracted proxy problem represented as a directed graph, in which vertices and edges are learned end-to-end from hindsight. Our theoretical analyses provide performance guarantees under appropriate assumptions and establish where our approach is expected to be helpful. Generalization-focused experiments validate Skipper’s significant advantage in zero-shot generalization, compared to some existing state-of-the-art hierarchical planning methods.

Decision Transformer-based decision-making agents have shown the ability to generalize across multiple tasks. However, their performance relies on massive data and computation. We argue that this inefficiency stems from the forgetting phenomenon, in which a model memorizes its behaviors in parameters throughout training. As a result, training on a new task may deteriorate the model’s performance on previous tasks. In contrast to LLMs’ implicit memory mechanism, the human brain utilizes distributed memory storage, which helps manage and organize multiple skills efficiently, mitigating the forgetting phenomenon. Inspired by this, we propose a working memory module to store, blend, and retrieve information for different downstream tasks. Evaluation results show that the proposed method improves training efficiency and generalization in Atari games and Meta-World object manipulation tasks. Moreover, we demonstrate that memory fine-tuning further enhances the adaptability of the proposed architecture.

The state-action occupancy measure of a policy is the expected (discounted or undiscounted) number of times a state-action couple is visited in a trajectory. For decades, RL books have been reporting the occupancy equivalence between Markovian and non-Markovian policies in countable state-action spaces under mild conditions. This equivalence states that the occupancy of any non-Markovian policy can be equivalently obtained by a Markovian policy, i.e. a memoryless probability distribution, conditioned only on its current state. While expected, for technical reasons, the translation of this result to continuous state space has resisted until now. Our main contribution is to fill this gap and to provide a general measure-theoretic treatment of the problem, permitting, in particular, its extension to continuous MDPs. Furthermore, we show that when the occupancy is infinite, we may encounter some non-trivial cases where the result does not hold anymore.

SARSA, a classical on-policy control algorithm for reinforcement learning, is known to chatter when combined with linear function approximation: SARSA does not diverge but oscillates in a bounded region. However, little is known about how fast SARSA converges to that region and how large the region is. In this paper, we make progress towards this open problem by showing the convergence rate of projected SARSA to a bounded region. Importantly, the region is much smaller than the region that we project into, provided that the the magnitude of the reward is not too large. Existing works regarding the convergence of linear SARSA to a fixed point all require the Lipschitz constant of SARSA’s policy improvement operator to be sufficiently small; our analysis instead applies to arbitrary Lipschitz constants and thus characterizes the behavior of linear SARSA for a new regime.

Offline reinforcement learning (RL) enables learning a decision-making policy without interaction with the environment. This makes it particularly beneficial in situations where such interactions are costly. However, a known challenge for offline RL algorithms is the distributional mismatch between the state-action distributions of the learned policy and the dataset, which can significantly impact performance. State-of-the-art algorithms address it by constraining the policy to align with the state-action pairs in the dataset. However, this strategy struggles on datasets that predominantly consist of trajectories collected by low-performing policies and only a few trajectories from high-performing ones. Indeed, the constraint to align with the data leads the policy to imitate low-performing behaviors predominating the dataset. Our key insight to address this issue is to constrain the policy to the policy that collected the good parts of the dataset rather than all data. To this end, we optimize the importance sampling weights to emulate sampling data from a data distribution generated by a nearly optimal policy. Our method exhibits considerable performance gains (up to five times better) over the existing approaches in state-of-the-art offline RL algorithms over 72 imbalanced datasets with varying types of imbalance.

While bisimulation-based approaches hold promise for learning robust state representations for Reinforcement Learning (RL) tasks, their efficacy in offline RL tasks has not been up to par. In some instances, their performance has even significantly underperformed alternative methods. We aim to understand why bisimulation methods succeed in online settings, but falter in offline tasks. Our analysis reveals that missing transitions in the dataset are particularly harmful to the bisimulation principle, leading to ineffective estimation. We also shed light on the critical role of reward scaling in bounding the scale of bisimulation measurements and of the value error they induce. Based on these findings, we propose to apply the expectile operator for representation learning to our offline RL setting, which helps to prevent overfitting to incomplete data. Meanwhile, by introducing an appropriate reward scaling strategy, we avoid the risk of feature collapse in representation space. We implement these recommendations on two state-of-the-art bisimulation-based algorithms, MICo and SimSR, and demonstrate performance gains on two benchmark suites: D4RL and Visual D4RL. Codes are provided at \url{https://github.com/zanghyu/Offline_Bisimulation}.

In Reinforcement Learning, the optimal action at a given state is dependent on policy decisions at subsequent states. As a consequence, the learning targets evolve with time and the policy optimization process must be efficient at unlearning what it previously learnt. In this paper, we discover that the policy gradient theorem prescribes policy updates that are slow to unlearn because of their structural symmetry with respect to the value target. To increase the unlearning speed, we study a novel policy update: the gradient of the cross-entropy loss with respect to the action maximizing $q$, but find that such updates may lead to a decrease in value. Consequently, we introduce a modified policy update devoid of that flaw, and prove its guarantees of convergence to global optimality in $\mathcal{O}(t^{-1})$ under classic assumptions. Further, we assess standard policy updates and our cross-entropy policy updates along six analytical dimensions. Finally, we empirically validate our theoretical findings.

Several recent works have proposed a class of algorithms for the offline reinforcement learning (RL) problem that we will refer to as return-conditioned supervised learning (RCSL). RCSL algorithms learn the distribution of actions conditioned on both the state and the return of the trajectory. Then they define a policy by conditioning on achieving high return. In this paper, we provide a rigorous study of the capabilities and limitations of RCSL something which is crucially missing in previous work. We find that RCSL returns the optimal policy under a set of assumptions that are stronger than those needed for the more traditional dynamic programming-based algorithms. We provide specific examples of MDPs and datasets that illustrate the necessity of these assumptions and the limits of RCSL. Finally, we present empirical evidence that these limitations will also cause issues in practice by providing illustrative experiments in simple point-mass environments and on datasets from the D4RL benchmark.

Goal-conditioned reinforcement learning (RL) is a promising direction for training agents that are capable of solving multiple tasks and reach a diverse set of objectives. How to \textit{specify} and \textit{ground} these goals in such a way that we can both reliably reach goals during training as well as generalize to new goals during evaluation remains an open area of research. Defining goals in the space of noisy, high-dimensional sensory inputs is one possibility, yet this poses a challenge for training goal-conditioned agents, or even for generalization to novel goals. We propose to address this by learning compositional representations of goals and processing the resulting representation via a discretization bottleneck, for coarser specification of goals, through an approach we call DGRL. We show that discretizing outputs from goal encoders through a bottleneck can work well in goal-conditioned RL setups, by experimentally evaluating this method on tasks ranging from maze environments to complex robotic navigation and manipulation tasks. Additionally, we show a theoretical result which bounds the expected return for goals not observed during training, while still allowing for specifying goals with expressive combinatorial structure.

In this paper, we establish the global optimality and convergence rate of an off-policy actor critic algorithm in the tabular setting without using density ratio to correct the discrepancy between the state distribution of the behavior policy and that of the target policy. Our work goes beyond existing works on the optimality of policy gradient methods in that existing works use the exact policy gradient for updating the policy parameters while we use an approximate and stochastic update step. Our update step is not a gradient update because we do not use a density ratio to correct the state distribution, which aligns well with what practitioners do. Our update is approximate because we use a learned critic instead of the true value function. Our update is stochastic because at each step the update is done for only the current state action pair. Moreover, we remove several restrictive assumptions from existing works in our analysis. Central to our work is the finite sample analysis of a generic stochastic approximation algorithm with time-inhomogeneous update operators on time-inhomogeneous Markov chains, based on its uniform contraction properties.

By the age of two, children tend to assume that new word categories are based on objects' shape, rather than their color or texture; this assumption is called the shape bias. They are thought to learn this bias by observing that their caregiver's language is biased towards shape based categories. This presents a chicken and egg problem: if the shape bias must be present in the language in order for children to learn it, how did it arise in language in the first place? In this paper, we propose that communicative efficiency explains both how the shape bias emerged and why it persists across generations. We model this process with neural emergent language agents that learn to communicate about raw pixelated images. First, we show that the shape bias emerges as a result of efficient communication strategies employed by agents. Second, we show that pressure brought on by communicative need is also necessary for it to persist across generations; simply having a shape bias in an agent's input language is insufficient. These results suggest that, over and above the operation of other learning strategies, the shape bias in human learners may emerge and be sustained by communicative pressures.

The policy gradient theorem states that the policy should only be updated in states that are visited by the current policy, which leads to insufficient planning in the off-policy states, and thus to convergence to suboptimal policies. We tackle this planning issue by extending the policy gradient theory to policy updates with respect to any state density. Under these generalized policy updates, we show convergence to optimality under a necessary and sufficient condition on the updates’ state densities, and thereby solve the aforementioned planning issue. We also prove asymptotic convergence rates that significantly improve those in the policy gradient literature. To implement the principles prescribed by our theory, we propose an agent, Dr Jekyll & Mr Hyde (J&H), with a double personality: Dr Jekyll purely exploits while Mr Hyde purely explores. J&H’s independent policies allow to record two separate replay buffers: one on-policy (Dr Jekyll’s) and one off-policy (Mr Hyde’s), and therefore to update J&H’s models with a mixture of on-policy and off-policy updates. More than an algorithm, J&H defines principles for actor-critic algorithms to satisfy the requirements we identify in our analysis. We extensively test on finite MDPs where J&H demonstrates a superior ability to recover from converging to a suboptimal policy without impairing its speed of convergence. We also implement a deep version of the algorithm and test it on a simple problem where it shows promising results.

We study the problem of Safe Policy Improvement (SPI) under constraints in the offline Reinforcement Learning (RL) setting. We consider the scenario where: (i) we have a dataset collected under a known baseline policy, (ii) multiple reward signals are received from the environment inducing as many objectives to optimize. We present an SPI formulation for this RL setting that takes into account the preferences of the algorithm’s user for handling the trade-offs for different reward signals while ensuring that the new policy performs at least as well as the baseline policy along each individual objective. We build on traditional SPI algorithms and propose a novel method based on Safe Policy Iteration with Baseline Bootstrapping (SPIBB, Laroche et al., 2019) that provides high probability guarantees on the performance of the agent in the true environment. We show the effectiveness of our method on a synthetic grid-world safety task as well as in a real-world critical care context to learn a policy for the administration of IV fluids and vasopressors to treat sepsis.

Malfunctioning neurons in the brain sometimes operate synchronously, reportedly causing many neurological diseases, e.g. Parkinson’s. Suppression and control of this collective synchronous activity are therefore of great importance for neuroscience, and can only rely on limited engineering trials due to the need to experiment with live human brains. We present the first Reinforcement Learning (RL) gym framework that emulates this collective behavior of neurons and allows us to find suppression parameters for the environment of synthetic degenerate models of neurons. We successfully suppress synchrony via RL for three pathological signaling regimes, characterize the framework’s stability to noise, and further remove the unwanted oscillations by engaging multiple PPO agents.

Playing text-based games requires skills in processing natural language and sequential decision making. Achieving human-level performance on text-based games remains an open challenge, and prior research has largely relied on hand-crafted structured representations and heuristics. In this work, we investigate how an agent can plan and generalize in text-based games using graph-structured representations learned end-to-end from raw text. We propose a novel graph-aided transformer agent (GATA) that infers and updates latent belief graphs during planning to enable effective action selection by capturing the underlying game dynamics. GATA is trained using a combination of reinforcement and self-supervised learning. Our work demonstrates that the learned graph-based representations help agents converge to better policies than their text-only counterparts and facilitate effective generalization across game configurations. Experiments on 500+ unique games from the TextWorld suite show that our best agent outperforms text-based baselines by an average of 24.2%.

We consider the decentralized exploration problem: a set of players collaborate to identify the best arm by asynchronously interacting with the same stochastic environment. The objective is to insure privacy in the best arm identification problem between asynchronous, collaborative, and thrifty players. In the context of a digital service, we advocate that this decentralized approach allows a good balance between conflicting interests: the providers optimize their services, while protecting privacy of users and saving resources. We define the privacy level as the amount of information an adversary could infer by intercepting all the messages concerning a single user. We provide a generic algorithm DECENTRALIZED ELIMINATION, which uses any best arm identification algorithm as a subroutine. We prove that this algorithm insures privacy, with a low communication cost, and that in comparison to the lower bound of the best arm identification problem, its sample complexity suffers from a penalty depending on the inverse of the probability of the most frequent players. Then, thanks to the genericity of the approach, we extend the proposed algorithm to the non-stationary bandits. Finally, experiments illustrate and complete the analysis.

This paper considers Safe Policy Improvement (SPI) in Batch Reinforcement Learning (Batch RL): from a fixed dataset and without direct access to the true environment, train a policy that is guaranteed to perform at least as well as the baseline policy used to collect the data. Our approach, called SPI with Baseline Bootstrapping (SPIBB), is inspired by the knows-what-it-knows paradigm: it bootstraps the trained policy with the baseline when the uncertainty is high. Our first algorithm, $\Pi_b$-SPIBB, comes with SPI theoretical guarantees. We also implement a variant, $\Pi_{\leq b}$-SPIBB, that is even more efficient in practice. We apply our algorithms to a motivational stochastic gridworld domain and further demonstrate on randomly generated MDPs the superiority of SPIBB with respect to existing algorithms, not only in safety but also in mean performance. Finally, we implement a model-free version of SPIBB and show its benefits on a navigation task with deep RL implementation called SPIBB-DQN, which is, to the best of our knowledge, the first RL algorithm relying on a neural network representation able to train efficiently and reliably from batch data, without any interaction with the environment.

A Budgeted Markov Decision Process (BMDP) is an extension of a Markov Decision Process to critical applications requiring safety constraints. It relies on a notion of risk implemented in the shape of an upper bound on a constrains violation signal that -- importantly -- can be modified in real-time. So far, BMDPs could only be solved in the case of finite state spaces with known dynamics. This work extends the state-of-the-art to continuous spaces environments and unknown dynamics. We show that the solution to a BMDP is the fixed point of a novel Budgeted Bellman Optimality operator. This observation allows us to introduce natural extensions of Deep Reinforcement Learning algorithms to address large-scale BMDPs. We validate our approach on two simulated applications: spoken dialogue and autonomous driving.

In Reinforcement Learning, an intelligent agent has to make a sequence of decisions to accomplish a goal. If this sequence is long, then the agent has to plan over a long horizon. While learning the optimal policy and its value function is a well studied problem in Reinforcement Learning, this paper focuses on the structure of the optimal value function and how hard it is to represent the optimal value function. We show that the generalized Rademacher complexity of the hypothesis space of all optimal value functions is dependent on the planning horizon and independent of the state and action space size. Further, we present bounds on the action-gaps of action value functions and show that they can collapse if a long planning horizon is used. The theoretical results are verified empirically on randomly generated MDPs and on a grid-world fruit collection task using deep value function approximation. Our theoretical results highlight a connection between value function approximation and the Options framework and suggest that value functions should be decomposed along bottlenecks of the MDP's transition dynamics.

This article addresses a particular Transfer Reinforcement Learning (RL) problem: when dynamics do not change from one task to another, and only the reward function does. Our method relies on two ideas, the first one is that transition samples obtained from a task can be reused to learn on any other task: an immediate reward estimator is learnt in a supervised fashion and for each sample, the reward entry is changed by its reward estimate. The second idea consists in adopting the optimism in the face of uncertainty principle and to use upper bound reward estimates. Our method is tested on a navigation task, under four Transfer RL experimental settings: with a known reward function, with strong and weak expert knowledge on the reward function, and with a completely unknown reward function. It is also evaluated in a Multi-Task RL experiment and compared with the state-of-the-art algorithms. Results reveal that this method constitutes a major improvement for transfer/multi-task problems that share dynamics.

One of the main challenges in reinforcement learning (RL) is generalisation. In typical deep RL methods this is achieved by approximating the optimal value function with a low-dimensional representation using a deep network. While this approach works well in many domains, in domains where the optimal value function cannot easily be reduced to a low-dimensional representation, learning can be very slow and unstable. This paper contributes towards tackling such challenging domains, by proposing a new method, called Hybrid Reward Architecture (HRA). HRA takes as input a decomposed reward function and learns a separate value function for each component reward function. Because each component typically only depends on a subset of all features, the corresponding value function can be approximated more easily by a low-dimensional representation, enabling more effective learning. We demonstrate HRA on a toy-problem and the Atari game Ms. Pac-Man, where HRA achieves above-human performance.

In this paper we introduce a novel model for computer-aided human-human dialogue. In this context, the computer aims at improving the outcome of a human-human task-oriented dialogue by intervening during the course of the interaction. While dialogue state and topic tracking in human-human dialogue have already been studied, few work has been devoted to the sequential part of the problem, where the impact of the system's actions on the future of the conversation is taken into account. This paper addresses this issue by first modelling human-human dialogue as a Markov Reward Process. The task of purposely taking part into the conversation is then optimised within the Linearly Solvable Markov Decision Process framework. Utterances of the Conversational Agent are seen as perturbations in this process, which aim at satisfying the user's long-term goals while keeping the conversation natural. Finally, results obtained by simulation suggest that such an approach is suitable for computer-aided human-human dialogue and is a first step towards three-party dialogue.

In this paper, we introduce a novel framework to encode the dynamics of dialogues into a probabilistic graphical model. Traditionally, Hidden Markov Models (HMMs) would be used to address this problem, involving a first step of hand-crafting to build a dialogue model (e.g. defining potential hidden states) followed by applying expectation-maximisation (EM) algorithms to refine it. Recently, an alternative class of algorithms based on the Method of Moments (MoM) has proven successful in avoiding issues of the EM-like algorithms such as convergence towards local optima, tractability issues, initialization issues or the lack of theoretical guarantees. In this work, we show that dialogues may be modeled by SP-RFA, a class of graphical models efficiently learnable within the MoM and directly usable in planning algorithms (such as reinforcement learning). Experiments are led on the Ubuntu corpus and dialogues are considered as sequences of dialogue acts, represented by their Latent Dirichlet Allocation (LDA) and Latent Semantic Analysis (LSA). We show that a MoM-based algorithm can learn a compact model of sequences of such acts.

This paper proposes a novel approach to defining and simulating a new generation of virtual personal assistants as multi-application multi-domain distributed dialogue systems. The first contribution is the assistant architecture, composed of independent third-party applications handled by a Dispatcher. In this view, applications are black-boxes responding with a self-scored answer to user requests. Next, the Dispatcher distributes the current request to the most relevant application, based on these scores and the context (history of interaction etc.), and conveys its answer to the user. To address variations in the user-defined portfolio of applications, the second contribution, a stochastic model automates the online optimisation of the Dispatcher's behaviour. To evaluate the learnability of the Dispatcher's policy, several parametrisations of the user and application simulators are enabled, in such a way that they cover variations of realistic situations. Results confirm in all considered configurations of interest, that reinforcement learning can learn adapted strategies.

In this paper, a turn-taking phenomenon taxonomy is introduced, organised according to the level of information conveyed.It is aimed to provide a better grasp of the behaviours used by humans while talking to each other, so that they can be methodically replicated in spoken dialogue systems.Five interesting phenomena have been implemented in a simulated environment: the system barge-in with three variants (resulting from either an unclear, an incoherent or a sufficient user message), the feedback and the user barge-in.The experiments reported in the paper illustrate that how such phenomena are implemented is a delicate choice as their impact on the system's performance is variable.

In this paper, an original framework to model human-machine spoken dialogues is proposed to deal with co-adaptation between users and Spoken Dialogue Systems in non-cooperative tasks.The conversation is modeled as a Stochastic Game: both the user and the system have their own preferences but have to come up with an agreement to solve a non-cooperative task.They are jointly trained so the Dialogue Manager learns the optimal strategy against the best possible user.Results obtained by simulation show that non-trivial strategies are learned and that this framework is suitable for dialogue modeling.

In this paper, reinforcement learning (RL) is used to learn an efficient turn-taking management model in a simulated slotfilling task with the objective of minimising the dialogue duration and maximising the completion task ratio.Turn-taking decisions are handled in a separate new module, the Scheduler.Unlike most dialogue systems, a dialogue turn is split into microturns and the Scheduler makes a decision for each one of them.A Fitted Value Iteration algorithm, Fitted-Q, with a linear state representation is used for learning the state to action policy.Comparison between a non-incremental and an incremental handcrafted strategies, taken as baselines, and an incremental RL-based strategy, shows the latter to be significantly more efficient, especially in noisy environments.

The automatic prediction of the quality of a dialogue is useful to keep track of a spoken dialogue system's performance and, if necessary, adapt its behaviour. Classifiers and regression models have been suggested to make this prediction. The parameters of these models are learnt from a corpus of dialogues evaluated by users or experts. In this paper, we propose to model this task as an ordinal regression problem. We apply support vector machines for ordinal regression on a corpus of dialogues where each system-user exchange was given a rate on a scale of 1 to 5 by experts. Compared to previous models proposed in the literature, the ordinal regression predictor has significantly better results according to the following evaluation metrics: Cohen's agreement rate with experts ratings, Spearman's rank correlation coefficient, and Euclidean and Manhattan errors.

Incrementality as a way of managing the interactions between a dialogue system and its users has been shown to have concrete advantages over the traditional turn-taking frame. Incremental systems are more reactive, more human-like, of-fer a better user experience and allow the user to correct errors faster, hence avoid-ing desynchronisations. Several incremen-tal models have been proposed, however, their core underlying architecture is dif-ferent from the classical dialogue systems. As a result, they have to be implemented from scratch. In this paper, we propose a method to transform traditional dialogue systems into incremental ones. A new module, called the Scheduler is inserted between the client and the service so that from the client’s point of view, the sys-tem behaves incrementally, even though the service does not. 1

In the Spoken Dialogue System literature, all studies consider the dialogue move as the unquestionable unit for reinforcement learning. Rather than learning at the dialogue move level, we apply the learning at the design level for three reasons : 1/ to alleviate the high-skill prerequisite for developers, 2/ to reduce the learning complexity by taking into account just the relevant subset of the context and 3/ to have interpretable learning results that carry a reusable usage feedback. Unfortunately, tackling the problem at the design level breaks the Markovian assumptions that are required in most Reinforcement Learning techniques. Consequently, we decided to use a recent non-Markovian algorithm called Compliance Based Reinforcement Learning. This paper presents the first experimentation on online optimisation in dialogue systems. It reveals a fast and significant improvement of the system performance with by average one system misunderstanding less per dialogue. Index Terms : Spoken Dialogue Systems, Reinforcement Learning, Online Learning, Hybrid System

This paper addresses the problem of introducing learning capabilities in industrial handcrafted automata-based Spoken Dialogue Systems, in order to help the developer to cope with his dialogue strategies design tasks. While classical reinforcement learning algorithms position their learning at the dialogue move level, the fundamental idea behind our approach is to learn at a finer internal decision level (which question, which words, which prosody, . . . ). These internal decisions are made on the basis of different (distinct or overlapping) knowledge. This paper proposes a novel reinforcement learning algorithm that can be used to make a data-driven optimisation of such handcrafted systems. An experiment shows that the convergence can be up to 20 times faster than with Q-Learning.