The recent rapid progress in (self) supervised learning models is in large part predicted by empirical scaling laws: a model's performance scales proportionally to its size. Analogous scaling laws remain elusive for reinforcement learning domains, however, where increasing the parameter count of a model often hurts its final performance. In this paper, we demonstrate that incorporating Mixture-of-Expert (MoE) modules, and in particular Soft MoEs (Puigcerver et al., 2023), into value-based networks results in more parameter-scalable models, evidenced by substantial performance increases across a variety of training regimes and model sizes. This work thus provides strong empirical evidence towards developing scaling laws for reinforcement learning.

Our work addresses a fundamental problem in the context of counterfactual inference for Markov Decision Processes (MDPs). Given an MDP path $\tau$, this kind of inference allows us to derive counterfactual paths $\tau'$ describing what-if versions of $\tau$ obtained under different action sequences than those observed in $\tau$. However, as the counterfactual states and actions deviate from the observed ones over time, the observation $\tau$ may no longer influence the counterfactual world, meaning that the analysis is no longer tailored to the individual observation, resulting in interventional outcomes rather than counterfactual ones. Even though this issue specifically affects the popular Gumbel-max structural causal model used for MDP counterfactuals, it has remained overlooked until now. In this work, we introduce a formal characterisation of influence based on comparing counterfactual and interventional distributions. We devise an algorithm to construct counterfactual models that automatically satisfy influence constraints. Leveraging such models, we derive counterfactual policies that are not just optimal for a given reward structure but also remain tailored to the observed path. Even though there is an unavoidable trade-off between policy optimality and strength of influence constraints, our experiments demonstrate that it is possible to derive (near-)optimal policies while remaining under the influence of the observation.

Autonomous vehicles have to obey traffic rules. These rules are often formalized using temporal logic, resulting in constraints that are hard to solve using optimization-based motion planners. Reinforcement Learning (RL) is a promising method to find motion plans adhering to temporal logic specifications. However, vanilla RL algorithms are based on random exploration, which is inherently unsafe. To address this issue, we propose a provably safe RL approach that always complies with traffic rules. As a specific application area, we consider vessels on the open sea, which must adhere to the Convention on the International Regulations for Preventing Collisions at Sea (COLREGS). We introduce an efficient verification approach that determines the compliance of actions with respect to the COLREGS formalized using temporal logic. Our action verification is integrated into the RL process so that the agent only selects verified actions. In contrast to agents that only integrate the traffic rule information in the reward function, our provably safe agent always complies with the formalized rules in critical maritime traffic situations and, thus, never causes a collision.

Digital twin (DT) platforms are increasingly regarded as a promising technology for controlling, optimizing, and monitoring complex engineering systems such as next-generation wireless networks. An important challenge in adopting DT solutions is their reliance on data collected offline, lacking direct access to the physical environment. This limitation is particularly severe in multi-agent systems, for which conventional multi-agent reinforcement (MARL) requires online interactions with the environment. A direct application of online MARL schemes to an offline setting would generally fail due to the epistemic uncertainty entailed by the limited availability of data. In this work, we propose an offline MARL scheme for DT-based wireless networks that integrates distributional RL and conservative Q-learning to address the environment's inherent aleatoric uncertainty and the epistemic uncertainty arising from limited data. To further exploit the offline data, we adapt the proposed scheme to the centralized training decentralized execution framework, allowing joint training of the agents' policies. The proposed MARL scheme, referred to as multi-agent conservative quantile regression (MA-CQR) addresses general risk-sensitive design criteria and is applied to the trajectory planning problem in drone networks, showcasing its advantages.

Many existing obstacle avoidance algorithms overlook the crucial balance between safety and agility, especially in environments of varying complexity. In our study, we introduce an obstacle avoidance pipeline based on reinforcement learning. This pipeline enables drones to adapt their flying speed according to the environmental complexity. Moreover, to improve the obstacle avoidance performance in cluttered environments, we propose a novel latent space. The latent space in this representation is explicitly trained to retain memory of previous depth map observations. Our findings confirm that varying speed leads to a superior balance of success rate and agility in cluttered environments. Additionally, our memory-augmented latent representation outperforms the latent representation commonly used in reinforcement learning. Finally, after minimal fine-tuning, we successfully deployed our network on a real drone for enhanced obstacle avoidance.

One-shot imitation is to learn a new task from a single demonstration, yet it is a challenging problem to adopt it for complex tasks with the high domain diversity inherent in a non-stationary environment. To tackle the problem, we explore the compositionality of complex tasks, and present a novel skill-based imitation learning framework enabling one-shot imitation and zero-shot adaptation; from a single demonstration for a complex unseen task, a semantic skill sequence is inferred and then each skill in the sequence is converted into an action sequence optimized for environmental hidden dynamics that can vary over time. Specifically, we leverage a vision-language model to learn a semantic skill set from offline video datasets, where each skill is represented on the vision-language embedding space, and adapt meta-learning with dynamics inference to enable zero-shot skill adaptation. We evaluate our framework with various one-shot imitation scenarios for extended multi-stage Meta-world tasks, showing its superiority in learning complex tasks, generalizing to dynamics changes, and extending to different demonstration conditions and modalities, compared to other baselines.

This paper contributes a new approach for distributional reinforcement learning which elucidates a clean separation of transition structure and reward in the learning process. Analogous to how the successor representation (SR) describes the expected consequences of behaving according to a given policy, our distributional successor measure (SM) describes the distributional consequences of this behaviour. We formulate the distributional SM as a distribution over distributions and provide theory connecting it with distributional and model-based reinforcement learning. Moreover, we propose an algorithm that learns the distributional SM from data by minimizing a two-level maximum mean discrepancy. Key to our method are a number of algorithmic techniques that are independently valuable for learning generative models of state. As an illustration of the usefulness of the distributional SM, we show that it enables zero-shot risk-sensitive policy evaluation in a way that was not previously possible.

Automated and autonomous industrial inspection is a longstanding research field, driven by the necessity to enhance safety and efficiency within industrial settings. In addressing this need, we introduce an autonomously navigating robotic system designed for comprehensive plant inspection. This innovative system comprises a robotic platform equipped with a diverse array of sensors integrated to facilitate the detection of various process and infrastructure parameters. These sensors encompass optical (LiDAR, Stereo, UV/IR/RGB cameras), olfactory (electronic nose), and acoustic (microphone array) capabilities, enabling the identification of factors such as methane leaks, flow rates, and infrastructural anomalies. The proposed system underwent individual evaluation at a wastewater treatment site within a chemical plant, providing a practical and challenging environment for testing. The evaluation process encompassed key aspects such as object detection, 3D localization, and path planning. Furthermore, specific evaluations were conducted for optical methane leak detection and localization, as well as acoustic assessments focusing on pump equipment and gas leak localization.

This paper introduces a new approach for continual planning and model learning in non-stationary stochastic environments expressed using relational representations. Such capabilities are essential for the deployment of sequential decision-making systems in the uncertain, constantly evolving real world. Working in such practical settings with unknown (and non-stationary) transition systems and changing tasks, the proposed framework models gaps in the agent's current state of knowledge and uses them to conduct focused, investigative explorations. Data collected using these explorations is used for learning generalizable probabilistic models for solving the current task despite continual changes in the environment dynamics. Empirical evaluations on several benchmark domains show that this approach significantly outperforms planning and RL baselines in terms of sample complexity in non-stationary settings. Theoretical results show that the system reverts to exhibit desirable convergence properties when stationarity holds.

Scripted agents have predominantly won the five previous iterations of the IEEE microRTS (µRTS) competitions hosted at CIG and CoG. Despite Deep Reinforcement Learning (DRL) algorithms making significant strides in real-time strategy (RTS) games, their adoption in this primarily academic competition has been limited due to the considerable training resources required and the complexity inherent in creating and debugging such agents. In a benchmark without performance constraints, RAISocketAI regularly defeated the two prior competition winners. This first competition-winning DRL submission can be a benchmark for future microRTS competitions and a starting point for future DRL research. Iteratively fine-tuning the base policy and transfer learning to specific maps were critical to RAISocketAI's winning performance. These strategies can be used to economically train future DRL agents. Further work in Imitation Learning using Behavior Cloning and fine-tuning these models with DRL has proven promising as an efficient way to bootstrap models with demonstrated, competitive behaviors. Deep reinforcement learning (DRL) has proven to be powerful at solving complex problems requiring several steps to achieve a goal, such as Atari games (Mnih et al., 2013), continuous control tasks (Lillicrap et al., 2016), and even real-time strategy (RTS) games like StarCraft II (Vinyals et al., 2019). The StarCraft II grandmaster agent AlphaStar was trained with thousands of CPUs and GPUs/TPUs for several weeks. RTS games are particularly challenging for DRL for several reasons: (1) the observation and action spaces are large and varied with different terrain and unit types; (2) each unit type can have different actions and abilities; (3) each action can control several units at once; (4) rewards are sparse (win, loss, or tie) and delayed by possibly several thousand timesteps; (5) winning requires combining tactical (micro) and strategic (macro) decisions; (6) actions must be taken in real-time (i.e., the game won't wait for the agent to take an action); (7) the agent might not have full visibility of the game state (i.e., fog of war); and (8) events in the game might be non-deterministic. It includes many aspects of RTS games, simplified: different unit types, unit-specific actions, terrain, resource collection and utilization to build units, and unit-to-unit combat where units have different strengths and weaknesses. The IEEE microRTS competitions have been hosted at the Conference on Games (CoG) nearly every year since 2019 and at the Conference on Computational Intelligence and Games (CIG) before that since 2017 (Ontañón et al., 2018).