While deep reinforcement learning has proven to be successful in solving control tasks, the ``black-box'' nature of an agent has received increasing concerns. We propose a prototype-based post-hoc \emph{policy explainer}, ProtoX, that explains a black-box agent by prototyping the agent's behaviors into scenarios, each represented by a prototypical state. When learning prototypes, ProtoX considers both visual similarity and scenario similarity. The latter is unique to the reinforcement learning context since it explains why the same action is taken in visually different states. To teach ProtoX about visual similarity, we pre-train an encoder using contrastive learning via self-supervised learning to recognize states as similar if they occur close together in time and receive the same action from the black-box agent.

We study the design of explicable reward functions for a reinforcement learning agent while guaranteeing that an optimal policy induced by the function belongs to a set of target policies. By being explicable, we seek to capture two properties: (a) informativeness so that the rewards speed up the agent's convergence, and (b) sparseness as a proxy for ease of interpretability of the rewards. The key challenge is that higher informativeness typically requires dense rewards for many learning tasks, and existing techniques do not allow one to balance these two properties appropriately. In this paper, we investigate the problem from the perspective of discrete optimization and introduce a novel framework, ExpRD, to design explicable reward functions. ExpRD builds upon an informativeness criterion that captures the (sub-)optimality of target policies at different time horizons in terms of actions taken from any given starting state. We provide a mathematical analysis of ExpRD, and show its connections to existing reward design techniques, including potential-based reward shaping. Experimental results on two navigation tasks demonstrate the effectiveness of ExpRD in designing explicable reward functions.

Nearly all state-of-the-art deep learning algorithms rely on error backpropagation, which is generally regarded as biologically implausible. An alternative way of training an artificial neural network is through treating each unit in the network as a reinforcement learning agent, and thus the network is considered as a team of agents. As such, all units can be trained by REINFORCE, a local learning rule modulated by a global signal that is more consistent with biologically observed forms of synaptic plasticity. Although this learning rule follows the gradient of return in expectation, it suffers from high variance and thus the low speed of learning, rendering it impractical to train deep networks. We therefore propose a novel algorithm called MAP propagation to reduce this variance significantly while retaining the local property of the learning rule. Experiments demonstrated that MAP propagation could solve common reinforcement learning tasks at a similar speed to backpropagation when applied to an actor-critic network. Our work thus allows for the broader application of teams of agents in deep reinforcement learning.

Designing reinforcement learning (RL) agents is typically a difficult process that requires numerous design iterations. Learning can fail for a multitude of reasons and standard RL methods provide too few tools to provide insight into the exact cause. In this paper, we show how to integrate \textit{value decomposition} into a broad class of actor-critic algorithms and use it to assist in the iterative agent-design process. Value decomposition separates a reward function into distinct components and learns value estimates for each. These value estimates provide insight into an agent's learning and decision-making process and enable new training methods to mitigate common problems.

Emergent symbolic representations are critical for enabling developmental learning agents to plan and generalize across tasks. In this work, we investigate whether large language models (LLMs) can translate human natural language instructions into the internal symbolic representations that emerge during hierarchical reinforcement learning. We apply a structured evaluation framework to measure the translation performance of commonly seen LLMs -- GPT, Claude, Deepseek and Grok -- across different internal symbolic partitions generated by a hierarchical reinforcement learning algorithm in the Ant Maze and Ant Fall environments. Our findings reveal that although LLMs demonstrate some ability to translate natural language into a symbolic representation of the environment dynamics, their performance is highly sensitive to partition granularity and task complexity. The results expose limitations in current LLMs capacity for representation alignment, highlighting the need for further research on robust alignment between language and internal agent representations.

Reinforcement learning agents in complex game environments often suffer from sparse rewards, training instability, and poor sample efficiency. This paper presents a hybrid training approach that combines offline imitation learning with online reinforcement learning for a 2D shooter game agent. We implement a multi-head neural network with separate outputs for behavioral cloning and Q-learning, unified by shared feature extraction layers with attention mechanisms. Initial experiments using pure deep Q-Networks exhibited significant instability, with agents frequently reverting to poor policies despite occasional good performance. To address this, we developed a hybrid methodology that begins with behavioral cloning on demonstration data from rule-based agents, then transitions to reinforcement learning. Our hybrid approach achieves consistently above 70% win rate against rule-based opponents, substantially outperforming pure reinforcement learning methods which showed high variance and frequent performance degradation. The multi-head architecture enables effective knowledge transfer between learning modes while maintaining training stability. Results demonstrate that combining demonstration-based initialization with reinforcement learning optimization provides a robust solution for developing game AI agents in complex multi-agent environments where pure exploration proves insufficient.

Markov Decision Processes (MDPs) are the most common model for decision making under uncertainty in the Machine Learning community. An MDP captures non-determinism, probabilistic uncertainty, and an explicit model of action. A Reinforcement Learning (RL) agent learns to act in an MDP by maximizing a utility function. This paper considers the problem of learning a decision policy that maximizes utility subject to satisfying a constraint expressed in deontic logic. In this setup, the utility captures the agent's mission - such as going quickly from A to B. The deontic formula represents (ethical, social, situational) constraints on how the agent might achieve its mission by prohibiting classes of behaviors. We use the logic of Expected Act Utilitarianism, a probabilistic stit logic that can be interpreted over controlled MDPs. We develop a variation on policy improvement, and show that it reaches a constrained local maximum of the mission utility. Given that in stit logic, an agent's duty is derived from value maximization, this can be seen as a way of acting to simultaneously maximize two value functions, one of which is implicit, in a bi-level structure. We illustrate these results with experiments on sample MDPs.

Reinforcement learning (RL) is an actively growing field that is seeing increased usage in real-world, safety-critical applications -- making it paramount to ensure the robustness of RL algorithms against adversarial attacks. In this work we explore a particularly stealthy form of training-time attacks against RL -- backdoor poisoning. Here the adversary intercepts the training of an RL agent with the goal of reliably inducing a particular action when the agent observes a pre-determined trigger at inference time. We uncover theoretical limitations of prior work by proving their inability to generalize across domains and MDPs. Motivated by this, we formulate a novel poisoning attack framework which interlinks the adversary's objectives with those of finding an optimal policy -- guaranteeing attack success in the limit.

As reinforcement learning agents become increasingly deployed in real-world scenarios, predicting future agent actions and events during deployment is important for facilitating better human-agent interaction and preventing catastrophic outcomes. This paper experimentally evaluates and compares the effectiveness of future action and event prediction for three types of RL agents: explicitly planning, implicitly planning, and non-planning. We employ two approaches: the inner state approach, which involves predicting based on the inner computations of the agents (e.g., plans or neuron activations), and a simulation-based approach, which involves unrolling the agent in a learned world model. Our results show that the plans of explicitly planning agents are significantly more informative for prediction than the neuron activations of the other types. Furthermore, using internal plans proves more robust to model quality compared to simulation-based approaches when predicting actions, while the results for event prediction are more mixed.

While deep reinforcement learning has proven to be successful in solving control tasks, the black-box'' nature of an agent has received increasing concerns. We propose a prototype-based post-hoc \emph{policy explainer}, ProtoX, that explains a black-box agent by prototyping the agent's behaviors into scenarios, each represented by a prototypical state. When learning prototypes, ProtoX considers both visual similarity and scenario similarity. The latter is unique to the reinforcement learning context since it explains why the same action is taken in visually different states. To teach ProtoX about visual similarity, we pre-train an encoder using contrastive learning via self-supervised learning to recognize states as similar if they occur close together in time and receive the same action from the black-box agent.