52 papers across 87 tags
Papers
A public catalog of research papers organized by topic.
Q-Learning
. ~-learning (Watkins, 1989) is a simpleway for agentsto learn how to act optimallyin controlledMarkovian domains. It amounts to an incremental method for dynamic programming which imposes limited computational demands. It works by successivelyimprovingits evaluationsof the quali...
Efficient Model Checking of Safety Properties
. We consider the problems of identifying LTL safety proper- ties and translating them to finite automata. We present an algorithm for constructing a finite automata recognising informative prefixes of LTL formulas based on [1]. The implementation also includes a procedure for de...
Transition Systems over Games
1. A game, which is essentially a bipartite graph describing the We describe a framework for game semantics combining opera- moves that can be made as play passes between the different tional and denotational accounts. A game is a bipartite graph of positions of the two players....
Shield Synthesis: Runtime Enforcement for Reactive Systems
. Scalability issues may prevent users from verifying critical proper- ties of a complex hardware design. In this situation, we propose to synthesize a “safety shield” that is attached to the design to enforce the properties at run time. Shield synthesis can succeed where model c...
Constrained Policy Optimization
In reinforcement learning (RL), agents learn to act by trial and error, gradually improving their performance at the For many applications of reinforcement learn- task as learning progresses. Recent work in deep RL as- ing it can be more convenient to specify both sumes that agen...
Value-Decomposition Networks for Cooperative Multi-Agent Learning
We study the problem of cooperative multi-agent reinforcement learning with a single joint reward signal. This class of learning problems is difficult because of the often large combined action and observation spaces. In the fully centralized and decentralized approaches, we find...
Safe Reinforcement Learning via Shielding
reward Environment Learning Agent Reinforcement learning algorithms discover policies that observation maximize reward, but do not necessarily guarantee safety dur- actions ing learning or execution phases. We introduce a new ap- proach to learn optimal policies while enforcing p...
Sound Value Iteration
. Computing reachability probabilities is at the heart of prob- abilistic model checking. All model checkers compute these probabilities in an iterative fashion using value iteration. This technique approximates a xed point from below by determining reachability probabilities for...
Symbolic Verification of GOLOG Programs with First-Order BDDs
loop: while (∃x.OnRobot(x)) do πx:Dish unload (x) endWhile; G OLOG is an agent language that allows defining behaviours πy:Room { goto(y); in terms of programs over actions defined in a Situation Cal- while (∃xDirty(x, y)) do culus action theory. Often it is vital to verify such...
Dealing with Non-Stationarity in Multi-Agent Deep Reinforcement Learning
policies during the training procedure. This non-stationarity stems from breaking the Markov assumption that governs most Recent developments in deep reinforcement learning single-agent RL algorithms. Since the transitions and rewards are concerned with creating decision-making a...
A Symbolic Approach to Safety LTL Synthesis
. Temporal synthesis is the automated design of a system that interacts with an environment, using the declarative specification of the system’s behavior. A popular language for providing such a specification is Linear Temporal Logic, or LTL. LTL synthesis in the general case has...
Too Many Cooks: Coordinating Multi-Agent Collaboration Through Inverse Planning
AAMAS 2020, May 9–13, Auckland, New Zealand Too Many Cooks: Coordinating Multi-agent Collaboration Through Inverse Planning∗ Extended Abstract Rose E. Wang∗ Sarah A. Wu∗ James A. Evans MIT MIT U. Chicago rewang@mit.edu sarahawu@mit.edu jevans@uchicago.edu Joshua B. Tenenbaum Davi...
Deception in Social Learning: A Multi-Agent Reinforcement Learning Perspective
Within the framework of Multi-Agent Reinforcement Learning, Social Learning is a new class of algorithms that enables agents to reshape the reward function of other agents with the goal of promoting cooperation and achieving higher global rewards in mixed-motive games. However, t...
Safe Multi-Agent Reinforcement Learning via Shielding
A recent work [1] developed a shielding framework for single- Multi-agent reinforcement learning (MARL) has been increasingly agent reinforcement learning (RL), which synthesizes a shield to used in a wide range of safety-critical applications, which require enforce the correctne...
Automata Cascades: Expressivity and Sample Complexity
Every automaton can be decomposed into a cascade of ba- sic automata. This is the Prime Decomposition Theorem by Krohn and Rhodes. We show that cascades allow for describ- ing the sample complexity of automata in terms of their com- ponents. In particular, we show that the sample...
Deep Reinforcement Learning for Multi-Agent Interaction
. The development of autonomous agents which can interact with other agents to accomplish a given task is a core area of research in artificial intelligence and machine learning. Towards this goal, the Autonomous Agents Research Group develops novel machine learning algorithms fo...
First-Order Game Logic and Modal Mu-Calculus
Three decades later, Parikh’s problem has been solved: This paper investigates first-order game logic and first-order Game logic is less expressive than the modal μ-calculus [2], modal μ-calculus, which extend their propositional modal because it embeds into the two-variable frag...
Reward Machines: Exploiting Reward Function Structure in Reinforcement Learning
Reinforcement learning (RL) methods usually treat reward functions as black boxes. As such, these methods must extensively interact with the environment in order to discover rewards and optimal policies. In most RL applications, however, users have to program the reward function...
The Surprising Effectiveness of PPO in Cooperative Multi-Agent Games
Proximal Policy Optimization (PPO) is a ubiquitous on-policy reinforcement learn- ing algorithm but is significantly less utilized than off-policy learning algorithms in multi-agent settings. This is often due to the belief that PPO is significantly less sample efficient than off...
Verifiable and Compositional Reinforcement Learning Systems
However, there remain significant barriers to their deploy- ment in safety-critical engineering applications. Autonomous We propose a framework for verifiable and compositional vehicles, manufacturing robotics, and power systems manage- reinforcement learning (RL) in which a coll...
Complexity of Safety and coSafety Fragments of Linear Temporal Logic
ones in such a way to satisfy the formula. LTL realizability Linear Temporal Logic (LTL) is the de-facto standard tempo- is 2EXPTIME-complete, on both infinite (Rosner 1992) and ral logic for system specification, whose foundational prop- finite (De Giacomo and Vardi 2015) traces...
Correct-by-Construction Reinforcement Learning of Cardiac Pacemakers from Duration Calculus Requirements
A common cardiac arrhythmia, known as heart block or AV block, is a condition where the electrical signal con- As the complexity of artificial pacemakers continues to grow, trolling the ventricular contractions are partially or com- the importance of capturing its functional corr...
Disentangling Interaction using Maximum Entropy Reinforcement Learning in Multi-Agent Systems
. Research on multi-agent interaction involving both mul- tiple artificial agents and humans is still in its infancy. Most recent ap- proaches have focused on environments with collaboration-focused human behavior, or providing only a small, defined set of situations. When deploy...
Persistence, Causation and Counterfactuals
: We present a novel theory of nonmonotonic temporal reasoning. It is related to approaches based on causation, the main di erence being that in our ap- proach causal relations take a context argument which indicates under what circumstances they should hold. Our theory can handl...
Temporal Logics
Pure-Past Action Masking
by (default or sampled) safe ones; action projection, e.g., (Cheng et al. 2019), where unsafe actions are replaced by We present Pure-Past Action Masking (PPAM), a lightweight approach to action masking for safe reinforcement learning. projecting them to the safe action space; an...
Responsibility in a Multi-Value Strategic Setting
. Responsibility is a key notion in multi-agent systems and in creating safe, reliable and ethical AI. In particular, the evaluation of choices based on responsibility is useful for making robustly good deci- sions in unpredictable domains. However, most previous work on respon-...
Safe Model-Based Multi-Agent Mean-Field Reinforcement Learning
Many applications, e.g., in shared mobility, require coordinating a large number of agents. Mean-field reinforcement learning ad- dresses the resulting scalability challenge by optimizing the policy of a representative agent interacting with the infinite population of identical a...
Simulating Counterfactuals
Counterfactual inference considers a hypothetical intervention in a parallel world that shares some evidence with the factual world. If the evidence specifies a conditional dis- tribution on a manifold, counterfactuals may be analytically intractable. We present an algorithm for...
Symbolic Solution of Emerson-Lei Games for Reactive Synthesis
. Emerson-Lei conditions have recently attracted attention due to both their succinctness and their favorable closure properties. In the current work, we show how infinite-duration games with Emerson-Lei objectives can be analyzed in two different ways. First, we show that the Zi...
Towards Guaranteed Safe AI: A Framework for Ensuring Robust and Reliable AI Systems
approaches aim to provide high-assurance quantitative guar- Ensuring that AI systems reliably and robustly antees about the safety of an AI system’s behaviour through avoid harmful or dangerous behaviours is a cru- the use of three core components — a formal safety specifi- cial...
An Extended Benchmarking of Multi-Agent Reinforcement Learning Algorithms in Complex Fully Cooperative Tasks
Benchmarking of Multi-Agent Reinforcement Learning Algorithms in Com- Multi-Agent Reinforcement Learning (MARL) has recently emerged plex Fully Cooperative Tasks. In Proc. of the 24th International Conference on Autonomous Agents and Multiagent Systems (AAMAS 2025), Detroit, Mich...
Compositional Shielding and Reinforcement Learning for Multi-Agent Systems
In this paper, we present the first MA shielding approach that is Deep reinforcement learning has emerged as a powerful tool for truly compositional, does not require online communication, and obtaining high-performance policies. However, the safety of these provides absolute saf...
Explainably Safe Reinforcement Learning
Trust in a decision-making system requires both safety guarantees and the ability to interpret and understand its behavior. This is particularly important for learned systems, whose decision-making processes are often highly opaque. Shielding is a prominent model-based technique...
Probabilistic Shielding for Safe Reinforcement Learning
at least some probability p. This framework comes from a probabilistic version of what is usually called the “safety” In real-life scenarios, a Reinforcement Learning (RL) agent aiming to maximise their reward, must often also behave in a fragment of Linear Temporal Logic (LTL) (...
ProSh: Probabilistic Shielding for Model-free Reinforcement Learning
not assume access to a simulator or environment abstraction, and Safety is a major concern in reinforcement learning (RL): we aim at is compatible with continuous environments. developing RL systems that not only perform optimally, but are also We also show that optimizing over s...
Successful Misunderstandings: Learning to Coordinate Without Being Understood
. The main approach to evaluating communication is by as- sessing how well it facilitates coordination. If two or more individuals can coordinate through communication, it is generally assumed that they un- derstand one another. We investigate this assumption in a signaling game...
An Algebraic Exposition of the Theory of Dyadic Morality
to be pro-social (Abdulhai et al. 2024). These different ap- arXiv:2605.16153v1 [cs.AI] 15 May 2026 proaches can be operationalized in guiding the behaviors of This paper provides an algebraic exposition of the theory AI agents with varying ease due to their inherent underly- of...
The Multi-Agent Off-Switch Game
the waiting agent off; if the agent isn’t turned off, it can proceed The off-switch game framework has been instrumental in under- to take the action. This elegant result suggests that uncertainty standing corrigibility — the property that AI agents should allow about human prefe...