Algorithmic recourse provides actions to individuals who have been adversely affected by automated decision-making and helps them achieve a desired outcome. Knowing the recourse, however, does not guarantee that users would implement it perfectly, either due to environmental variability or personal choices. Recourse generation should thus anticipate its sub-optimal or noisy implementation. While several approaches have constructed recourse that accounts for robustness to small perturbation (i.e., noisy recourse implementation), they assume an entire recourse to be implemented in a single step and thus apply one-off uniform noise to it. Such assumption is unrealistic since recourse often includes multiple sequential steps which becomes harder to implement and subject to more noise. In this work, we consider recourse under plausible noise that adapts to the local data geometry and accumulates at every step of the way. We frame this problem as a Markov Decision Process and demonstrate that the distribution of our plausible noise satisfies the Markov property. We then propose the RObust SEquential (ROSE) recourse generator to output a sequence of steps that will lead to the desired outcome even under imperfect implementation. Given our plausible modelling of sub-optimal human actions and greater recourse robustness to accumulated uncertainty, ROSE can grant users higher chances of success under low recourse costs. Empirical evaluation shows our algorithm manages the inherent trade-off between recourse robustness and costs more effectively while ensuring its low sparsity and fast computation.

Dynamic and interactive traffic scenarios pose significant challenges for autonomous driving systems. Reinforcement learning (RL) offers a promising approach by enabling the exploration of driving policies beyond the constraints of pre-collected datasets and predefined conditions, particularly in complex environments. However, a critical challenge lies in effectively extracting spatial and temporal features from sequences of high-dimensional, multi-modal observations while minimizing the accumulation of errors over time. Additionally, efficiently guiding large-scale RL models to converge on optimal driving policies without frequent failures during the training process remains tricky. We propose an end-to-end model-based RL algorithm named Ramble to address these issues. Ramble processes multi-view RGB images and LiDAR point clouds into low-dimensional latent features to capture the context of traffic scenarios at each time step. A transformer-based architecture is then employed to model temporal dependencies and predict future states. By learning a dynamics model of the environment, Ramble can foresee upcoming traffic events and make more informed, strategic decisions. Our implementation demonstrates that prior experience in feature extraction and decision-making plays a pivotal role in accelerating the convergence of RL models toward optimal driving policies. Ramble achieves state-of-the-art performance regarding route completion rate and driving score on the CARLA Leaderboard 2.0, showcasing its effectiveness in managing complex and dynamic traffic situations.

Multi-objective reinforcement learning (MORL) excels at handling rapidly changing preferences in tasks that involve multiple criteria, even for unseen preferences. However, previous dominating MORL methods typically generate a fixed policy set or preference-conditioned policy through multiple training iterations exclusively for sampled preference vectors, and cannot ensure the efficient discovery of the Pareto front. Furthermore, integrating preferences into the input of policy or value functions presents scalability challenges, in particular as the dimension of the state and preference space grow, which can complicate the learning process and hinder the algorithm's performance on more complex tasks. To address these issues, we propose a two-stage Pareto front discovery algorithm called Constrained MORL (C-MORL), which serves as a seamless bridge between constrained policy optimization and MORL. Concretely, a set of policies is trained in parallel in the initialization stage, with each optimized towards its individual preference over the multiple objectives. Then, to fill the remaining vacancies in the Pareto front, the constrained optimization steps are employed to maximize one objective while constraining the other objectives to exceed a predefined threshold. Empirically, compared to recent advancements in MORL methods, our algorithm achieves more consistent and superior performances in terms of hypervolume, expected utility, and sparsity on both discrete and continuous control tasks, especially with numerous objectives (up to nine objectives in our experiments). In many real-world control and planning problems, multiple and sometimes even conflicting objectives are getting involved. For instance, in industrial control scenarios (Salvendy, 2001; Wang et al., 2023), maximizing utility and minimizing energy consumption are of particular interest as objectives to be optimized. Since different decision makers have heterogeneous preferences over these objectives, there may exist multiple Pareto-optimal policies (Roijers et al., 2014). Classical reinforcement learning (RL) methods typically involve training individual policies exclusively to align with each preference weight vector over multiple rewards (Nagabandi et al., 2018; Gupta et al., 2018). Yet it may lead to an enormous computational burden due to the overly dependence on the model retraining and fine-tuning stages. Moreover, such policies are hard to directly generalize or transfer to newer tasks (Cobbe et al., 2019; Taiga et al., 2022).

Policy evaluation estimates the performance of a policy by (1) collecting data from the environment and (2) processing raw data into a meaningful estimate. Due to the sequential nature of reinforcement learning, any improper data-collecting policy or data-processing method substantially deteriorates the variance of evaluation results over long time steps. Thus, policy evaluation often suffers from large variance and requires massive data to achieve the desired accuracy. In this work, we design an optimal combination of data-collecting policy and data-processing baseline. Theoretically, we prove our doubly optimal policy evaluation method is unbiased and guaranteed to have lower variance than previously best-performing methods. Empirically, compared with previous works, we show our method reduces variance substantially and achieves superior empirical performance.

Tabular reinforcement learning methods cannot operate directly on continuous state spaces. One solution for this problem is to partition the state space. A good partitioning enables generalization during learning and more efficient exploitation of prior experiences. Consequently, the learning process becomes faster and produces more reliable policies. However, partitioning introduces approximation, which is particularly harmful in the presence of nonlinear relations between state components. An ideal partition should be as coarse as possible, while capturing the key structure of the state space for the given problem. This work extracts partitions from the environment dynamics by symbolic execution. We show that symbolic partitioning improves state space coverage with respect to environmental behavior and allows reinforcement learning to perform better for sparse rewards. We evaluate symbolic state space partitioning with respect to precision, scalability, learning agent performance and state space coverage for the learnt policies.

We consider a two-player sequential game in which agents have the same reward function but may disagree on the transition probabilities of an underlying Markovian model of the world. By committing to play a specific policy, the agent with the correct model can steer the behavior of the other agent, and seek to improve utility. We model this setting as a multi-view decision process, which we use to formally analyze the positive effect of steering policies. Furthermore, we develop an algorithm for computing the agents' achievable joint policy, and we experimentally show that it can lead to a large utility increase when the agents' models diverge.

We introduce a new formulation of the Hidden Parameter Markov Decision Process (HiP-MDP), a framework for modeling families of related tasks using lowdimensional latent embeddings.

Deep reinforcement learning algorithms have been shown to learn complex tasks using highly general policy classes. However, sparse reward problems remain a significant challenge. Exploration methods based on novelty detection have been particularly successful in such settings but typically require generative or predictive models of the observations, which can be difficult to train when the observations are very high-dimensional and complex, as in the case of raw images. We propose a novelty detection algorithm for exploration that is based entirely on discriminatively trained exemplar models, where classifiers are trained to discriminate each visited state against all others. Intuitively, novel states are easier to distinguish against other states seen during training. We show that this kind of discriminative modeling corresponds to implicit density estimation, and that it can be combined with countbased exploration to produce competitive results on a range of popular benchmark tasks, including state-of-the-art results on challenging egocentric observations in the vizDoom benchmark.

Statistical performance bounds for reinforcement learning (RL) algorithms can be critical for high-stakes applications like healthcare. This paper introduces a new framework for theoretically measuring the performance of such algorithms called Uniform-PAC, which is a strengthening of the classical Probably Approximately Correct (PAC) framework. In contrast to the PAC framework, the uniform version may be used to derive high probability regret guarantees and so forms a bridge between the two setups that has been missing in the literature. We demonstrate the benefits of the new framework for finite-state episodic MDPs with a new algorithm that is Uniform-PAC and simultaneously achieves optimal regret and PAC guarantees except for a factor of the horizon.

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.