Author name cluster

Thibault Lahire

Possible papers associated with this exact author name in Arrow. This page groups case-insensitive exact name matches and is not a full identity disambiguation profile.

2 papers

2 author rows

AAMAS Conference 2026 Conference Paper

A Survey of Reinforcement Learning for Autonomous Air Combat: Current Progresses and Limitations

Alex Pierron
Thibault Lahire

Autonomous air combat represents one of the most demanding challenges in artificial intelligence, requiring agents to operate under uncertainty, partial observability, and adversarial dynamics. Reinforcement Learning and Multi-Agent Reinforcement Learning have recently emerged as promising approaches for enabling adaptive decision-making and coordination in this domain. This survey provides a structured overview of Reinforcement Learningpowered autonomous air combat, with emphasis on open-source environments, algorithmic frameworks, and hierarchical control architectures. We systematically compare three aspects: (1) singleagent and multi-agent settings, (2) full-control and hierarchical abstractions, and (3) the treatment of sensors and observability. Furthermore, we analyze the reproducibility of recent contributions, highlighting the tension between fidelity and accessibility across open and closed-source platforms. Beyond a methodological review, we identify persistent challenges related to scalability, transfer to real-world platforms, non symmetrical scenarios robustness, and computational requirements. By consolidating these advances and limitations, this survey aims to clarify the current state of the field, highlightopenproblems, andoutlinepathwaystowardmorerobust, scalable, and operationally relevant autonomous collaboration in future air combat systems.

PDF

ICML Conference 2022 Conference Paper

Large Batch Experience Replay

Thibault Lahire
Matthieu Geist
Emmanuel Rachelson

Several algorithms have been proposed to sample non-uniformly the replay buffer of deep Reinforcement Learning (RL) agents to speed-up learning, but very few theoretical foundations of these sampling schemes have been provided. Among others, Prioritized Experience Replay appears as a hyperparameter sensitive heuristic, even though it can provide good performance. In this work, we cast the replay buffer sampling problem as an importance sampling one for estimating the gradient. This allows deriving the theoretically optimal sampling distribution, yielding the best theoretical convergence speed. Elaborating on the knowledge of the ideal sampling scheme, we exhibit new theoretical foundations of Prioritized Experience Replay. The optimal sampling distribution being intractable, we make several approximations providing good results in practice and introduce, among others, LaBER (Large Batch Experience Replay), an easy-to-code and efficient method for sampling the replay buffer. LaBER, which can be combined with Deep Q-Networks, distributional RL agents or actor-critic methods, yields improved performance over a diverse range of Atari games and PyBullet environments, compared to the base agent it is implemented on and to other prioritization schemes.

Details