We present PORTAL, a novel framework for developing artificial intelligence agents capable of playing thousands of 3D video games through language-guided policy generation. By transforming decision-making problems into language modeling tasks, our approach leverages large language models (LLMs) to generate behavior trees represented in domain-specific language (DSL). This method eliminates the computational burden associated with traditional reinforcement learning approaches while preserving strategic depth and rapid adaptability. Our framework introduces a hybrid policy structure that combines rule-based nodes with neural network components, enabling both high-level strategic reasoning and precise low-level control. A dual-feedback mechanism incorporating quantitative game metrics and vision-language model analysis facilitates iterative policy improvement at both tactical and strategic levels. The resulting policies are instantaneously deployable, human-interpretable, and capable of generalizing across diverse gaming environments. Experimental results demonstrate PORTAL's effectiveness across thousands of first-person shooter (FPS) games, showcasing significant improvements in development efficiency, policy generalization, and behavior diversity compared to traditional approaches. PORTAL represents a significant advancement in game AI development, offering a practical solution for creating sophisticated agents that can operate across thousands of commercial video games with minimal development overhead. Experiment results on the 3D video games are best viewed on https://zhongwen.one/projects/portal .

In this paper, we propose a new solution to reward adaptation (RA), the problem where the learning agent adapts to a target reward function based on one or multiple existing behaviors learned a priori under the same domain dynamics but different reward functions. Learning the target behavior from scratch is possible but often inefficient given the available source behaviors. Our work represents a new approach to RA via the manipulation of Q-functions. Assuming that the target reward function is a known function of the source reward functions, our approach to RA computes bounds of the Q function. We introduce an iterative process to tighten the bounds, similar to value iteration. This enables action pruning in the target domain before learning even starts. We refer to such a method as Q-Manipulation (Q-M). We formally prove that our pruning strategy does not affect the optimality of the returned policy while empirically show that it improves the sample complexity. Q-M is evaluated in a variety of synthetic and simulation domains to demonstrate its effectiveness, generalizability, and practicality.

Reinforcement learning control algorithms face significant challenges due to out-of-distribution and inefficient exploration problems. While model-based reinforcement learning enhances the agent's reasoning and planning capabilities by constructing virtual environments, training such virtual environments can be very complex. In order to build an efficient inference model and enhance the representativeness of learning data, we propose the Counterfactual Experience Augmentation (CEA) algorithm. CEA leverages variational autoencoders to model the dynamic patterns of state transitions and introduces randomness to model non-stationarity. This approach focuses on expanding the learning data in the experience pool through counterfactual inference and performs exceptionally well in environments that follow the bisimulation assumption. Environments with bisimulation properties are usually represented by discrete observation and action spaces, we propose a sampling method based on maximum kernel density estimation entropy to extend CEA to various environments. By providing reward signals for counterfactual state transitions based on real information, CEA constructs a complete counterfactual experience to alleviate the out-of-distribution problem of the learning data, and outperforms general SOTA algorithms in environments with difference properties. Finally, we discuss the similarities, differences and properties of generated counterfactual experiences and real experiences. The code is available at https://github.com/Aegis1863/CEA.

Reinforcement learning (RL) has increasingly been applied to solve real-world planning problems, with progress in handling large state spaces and time horizons. However, a key bottleneck in many domains is that RL methods cannot accommodate large, combinatorially structured action spaces. In such settings, even representing the set of feasible actions at a single step may require a complex discrete optimization formulation. We leverage recent advances in embedding trained neural networks into optimization problems to propose SEQUOIA, an RL algorithm that directly optimizes for long-term reward over the feasible action space. Our approach embeds a Q-network into a mixed-integer program to select a combinatorial action in each timestep. Here, we focus on planning over restless bandits, a class of planning problems which capture many real-world examples of sequential decision making. RMAB, a broader class of restless bandits with combinatorial actions that cannot be decoupled across the arms of the restless bandit, requiring direct solving over the joint, exponentially large action space. Our approach significantly outperforms existing methods--which cannot address sequential planning and combinatorial selection simultaneously--by an average of 24.8% on these difficult instances. Reinforcement learning (RL) has made tremendous progress in recent years to solve a wide range of practical problems (Treloar et al., 2020; Marot et al., 2021; Silvestro et al., 2022; Degrave et al., 2022). While successful at dealing with large or infinite state spaces, RL struggles with discrete, combinatorial action spaces. This limitation is pertinent to many real-world sequential decisionmaking problems, where resource constraints frequently lead to combinatorial action spaces (Dulac-Arnold et al., 2020). Consider, for example, a sequential resource allocation problem in which public health workers are dispatched to visit patients. The workers each have a limited daily budget to maximize patient well-being. These requirements give rise to an exponentially large combinatorial action space to optimize over, even when the number of workers and patients is small.

In real-life scenarios, a Reinforcement Learning (RL) agent aiming to maximise their reward, must often also behave in a safe manner, including at training time. Thus, much attention in recent years has been given to Safe RL, where an agent aims to learn an optimal policy among all policies that satisfy a given safety constraint. However, strict safety guarantees are often provided through approaches based on linear programming, and thus have limited scaling. In this paper we present a new, scalable method, which enjoys strict formal guarantees for Safe RL, in the case where the safety dynamics of the Markov Decision Process (MDP) are known, and safety is defined as an undiscounted probabilistic avoidance property. Our approach is based on state-augmentation of the MDP, and on the design of a shield that restricts the actions available to the agent. We show that our approach provides a strict formal safety guarantee that the agent stays safe at training and test time. Furthermore, we demonstrate that our approach is viable in practice through experimental evaluation.

This study presents an emotion-aware navigation framework -- EmoBipedNav -- using deep reinforcement learning (DRL) for bipedal robots walking in socially interactive environments. The inherent locomotion constraints of bipedal robots challenge their safe maneuvering capabilities in dynamic environments. When combined with the intricacies of social environments, including pedestrian interactions and social cues, such as emotions, these challenges become even more pronounced. To address these coupled problems, we propose a two-stage pipeline that considers both bipedal locomotion constraints and complex social environments. Specifically, social navigation scenarios are represented using sequential LiDAR grid maps (LGMs), from which we extract latent features, including collision regions, emotion-related discomfort zones, social interactions, and the spatio-temporal dynamics of evolving environments. The extracted features are directly mapped to the actions of reduced-order models (ROMs) through a DRL architecture. Furthermore, the proposed framework incorporates full-order dynamics and locomotion constraints during training, effectively accounting for tracking errors and restrictions of the locomotion controller while planning the trajectory with ROMs. Comprehensive experiments demonstrate that our approach exceeds both model-based planners and DRL-based baselines. The hardware videos and open-source code are available at https://gatech-lidar.github.io/emobipednav.github.io/.

In industrial X-ray Computed Tomography (CT), the need for rapid in-line inspection is critical. Sparse-angle tomography plays a significant role in this by reducing the required number of projections, thereby accelerating processing and conserving resources. Most existing methods aim to balance reconstruction quality and scanning time, typically relying on fixed scan durations. Adaptive adjustment of the number of angles is essential; for instance, more angles may be required for objects with complex geometries or noisier projections. The concept of optimal stopping, which dynamically adjusts this balance according to varying industrial needs, remains underutilized. Building on our previous work, we integrate optimal stopping into sequential Optimal Experimental Design (OED). We propose a novel method for computing the policy gradient within the Actor-Critic framework, enabling the development of adaptive policies for informative angle selection and scan termination. Additionally, we investigated the gap between simulation and real-world applications in the context of the developed learning-based method. Our trained model, developed using synthetic data, demonstrates reliable performance when applied to real-world data. This approach enhances the flexibility of CT operations and expands the applicability of sparse-angle tomography in industrial settings.

Embedded systems power many modern applications and must often meet strict reliability, real-time, thermal, and power requirements. Task replication can improve reliability by duplicating a task's execution to handle transient and permanent faults, but blindly applying replication often leads to excessive overhead and higher temperatures. Existing design-time methods typically choose the number of replicas based on worst-case conditions, which can waste resources under normal operation. In this paper, we present RL-TIME, a reinforcement learning-based approach that dynamically decides the number of replicas according to actual system conditions. By considering both the reliability target and a core-level Thermal Safe Power (TSP) constraint at run-time, RL-TIME adapts the replication strategy to avoid unnecessary overhead and overheating. Experimental results show that, compared to state-of-the-art methods, RL-TIME reduces power consumption by 63%, increases schedulability by 53%, and respects TSP 72% more often.

Collaborative underwater target hunting, facilitated by multiple autonomous underwater vehicles (AUVs), plays a significant role in various domains, especially military missions. Existing research predominantly focuses on designing efficient and high-success-rate hunting policy, particularly addressing the target's evasion capabilities. However, in real-world scenarios, the target can not only adjust its evasion policy based on its observations and predictions but also possess eavesdropping capabilities. If communication among hunter AUVs, such as hunting policy exchanges, is intercepted by the target, it can adapt its escape policy accordingly, significantly reducing the success rate of the hunting mission. To address this challenge, we propose a covert communication-guaranteed collaborative target hunting framework, which ensures efficient hunting in complex underwater environments while defending against the target's eavesdropping. To the best of our knowledge, this is the first study to incorporate the confidentiality of inter-agent communication into the design of target hunting policy. Furthermore, given the complexity of coordinating multiple AUVs in dynamic and unpredictable environments, we propose an adaptive multi-agent diffusion policy (AMADP), which incorporates the strong generative ability of diffusion models into the multi-agent reinforcement learning (MARL) algorithm. Experimental results demonstrate that AMADP achieves faster convergence and higher hunting success rates while maintaining covertness constraints.

Open-Ended Learning (OEL) autonomous robots can acquire new skills and knowledge through direct interaction with their environment, relying on mechanisms such as intrinsic motivations and self-generated goals to guide learning processes. OEL robots are highly relevant for applications as they can autonomously leverage acquired knowledge to perform tasks beneficial to human users in unstructured environments, addressing challenges unforeseen at design time. However, OEL robots face a significant limitation: their openness may lead them to waste time learning information that is irrelevant to tasks desired by specific users. Here, we propose a solution called `Purpose-Directed Open-Ended Learning' (POEL), based on the novel concept of `purpose' introduced in previous work. A purpose specifies what users want the robot to achieve. The key insight of this work is that purpose can focus OEL on learning self-generated classes of tasks that, while unknown during autonomous learning (as typical in OEL), involve objects relevant to the purpose. This concept is operationalised in a novel robot architecture capable of receiving a human purpose through speech-to-text, analysing the scene to identify objects, and using a Large Language Model to reason about which objects are purpose-relevant. These objects are then used to bias OEL exploration towards their spatial proximity and to self-generate rewards that favour interactions with them. The solution is tested in a simulated scenario where a camera-arm-gripper robot interacts freely with purpose-related and distractor objects. For the first time, the results demonstrate the potential advantages of purpose-focused OEL over state-of-the-art OEL methods, enabling robots to handle unstructured environments while steering their learning toward knowledge acquisition relevant to users.