Generalization in Reinforcement Learning from Logical Specifications

Declarative Specifications for Efficient and Safe Reinforcement Learning

In recent years, there have been several developments combining reinforcement learning (RL) with techniques from theoretical computer science fields such as logic and formal methods. The main goal of these works was to improve training speed and quality, and in some cases also enforce safety constraints. In this dissertation, we present several works that followed this research line. First, we explore research directions concerning reward machines (RMs), an approach proposed to improve training speed and train agents in achieving tasks that require temporally extended behaviours. Given an abstraction of the environment in which the agent acts, we show how we can generate a reward machine from the set of all plans to achieve the task in the abstraction. As the plans come from an abstraction of the environment, the agent still needs to learn how to enact them in order to achieve the task, which is done via RL. Then, we synthesise reward machines in a cooperative multi-agent scenario by using Alternating-time Temporal Logic (ATL) formulas encoding coalition tasks. By model checking the ATL formula, we can obtain a strategy (if there is any) for the coalition to achieve the task, which is then translated to a RM and used to train the agents. We then present an extension of reward machines that endows them with a pushdown stack, obtaining a "pushdown reward machine" (pdRM). As pdRMs are based on pushdown automata, they can encode a strictly larger set of tasks compared to standard RMs, while still enabling more efficient learning compared to other approaches. Finally, we present a work in safe RL, where agents must also respect safety constraints. We present how to enforce safety constraints using pure-past linear-time temporal logic (PPLTL). Each action is associated to a PPLTL formula, and by evaluating the formulas at each timestep we determine which actions the agent can to perform, guaranteeing constraint satisfaction.

Constrained Reinforcement Learning Using Successor Representations

Real-world Reinforcement Learning depends on the ability to formulate safety constraints into a policy. A common way to model such constraints is to introduce an additional cost signal in the Markov Decision Process, which notifies the agent of unwanted behavior independently of the reward signal. Unfortunately, current methods are hard to adapt to changes in the cost function introduced by, e.g., domain shift or obstacles moving over time. The lack of adaptability means that policies are too unflexible to deal with complex real-world conditions. We propose the Safe Deep Successor Representation (SafeDSR), a novel method that allows quick retraining of policies towards new cost structures. SafeDSR extends the Deep Successor Representation (Kulkarni et al., 2016) to Constrained Reinforcement Learning by introducing a single learnable weight matrix to decouple the learned value function across dynamics, rewards, and costs. This matrix can be updated in a supervised manner instead of having to adapt the whole network if the cost structure of the environment changes. We demonstrate this ability in a freely configurable two-dimensional navigation environment and show that our method is competitive on a simple navigation task while being considerably more flexible

Learning Generalizable Behaviors for Terminal Agents

Terminal agents are a compelling application of large language models (LLMs), with the potential to integrate deeply into users'daily workflows. Reinforcement learning (RL) is a key technique for improving their capabilities, making scalable training environments a central challenge. Since public real-user interaction data are scarce, synthetic environments provide a practical alternative, but often suffer from domain gaps and limited fidelity, leading to poor generalization. Existing work mainly scales the quantity and diversity of synthetic environments, while reward-signal quality and the mechanisms governing generalization remain under-explored. We study how RL improves terminal agents and propose the Agentic Compositional Generalization hypothesis: rather than teaching new domain-specific skills from scratch, RL primarily shapes high-level decision-making behaviors that compose and route low-level skills acquired during pre-training and supervised fine-tuning (SFT). This account is consistent with our empirical results and suggests that verifier quality, which determines which behaviors are reinforced, is more important than simply increasing environment quantity or diversity. Motivated by this insight, we propose River, a simple training recipe that improves reward quality by filtering low-quality environments and augmenting outcome rewards with process-level behavior regularization. Using this recipe, our RL-trained agent achieves the best performance among evaluated open-source RL-trained 8B models across four terminal-agent benchmarks. River also generalizes across model families, scales, agent harnesses, and RL objectives. Using fewer than 30% of the TMax training environments, River improves RL gains by 106% and 30% on average for models ranging from 2B to 27B on Terminal-Bench-Lite and Terminal-Bench-v2.1, respectively.

Cross-Benchmark Generalization in Long-Horizon Agents

For reinforcement learning (RL) in self-contained environments, a policy can get rewards by exploiting environment-specific regularities (tool schemas, grader parsing, task templates) rather than by acquiring transferable skill, and an in-distribution holdout shares those regularities. We argue that the discriminating question is behavioral, namely how a trained agent acts, and that cross-benchmark transfer is the right place to look for it. We post-train an open-weight mixture-of-experts model (Qwen3.5-122B-A10B) on 363 long-horizon Model Context Protocol (MCP) tasks across 27 categories, using a two-stage SFT-then-RL pipeline. Toolathlon performance informed the initial base-family and SFT-teacher choices, but no external-benchmark task or grader entered training and no external score informed the reward, training hyperparameters, trained-checkpoint selection, or stopping. At greedy pass@1, the trained model improves over the base on five reported external evaluations: Toolathlon (+9.6 pp), $\tau^2$-Bench (+5.3 pp), BFCL-V4 (+3.5 pp), SWE-Bench Pro (+5.8 pp), and Terminal-Bench 2 (+2.8 pp). Both software-engineering benchmarks improve despite the training collection containing no software-engineering tasks. An exploratory paired-trajectory analysis identifies four recurring behavioral differences (more careful local-goal formation, building goal-relevant working state, keeping parent goals stable through local repairs, and verifying completion) that appear in analogous forms across office workflows and code. These results provide descriptive evidence that long-horizon multi-tool post-training can change ways of working that transfer beyond its training domain.

A Simple Baseline for Learning Approximate State Abstractions in Factored State Spaces

From Trajectories to Instructions: Language-Conditioned Meta-Reinforcement Learning

Model-Agnostic Meta-Learning (MAML) is a widely used framework for reinforcement learning (RL) that enables efficient transfer by learning global policy parameters that can be rapidly adapted to new tasks. MAML training proceeds in two loops: an inner loop where the global parameters are adapted to task-specific parameters, and an outer loop where these task-specific parameters are evaluated and losses are back-propagated to improve the global parameters. Traditionally, the inner loop adaptation is performed by collecting trajectories from the task environment and applying gradient updates on the empirical expected return, which can be a costly operation. We note that it is the outer loop that drives the actual learning of global parameters, and therefore the inner loop adaptation mechanism need not be restricted to be gradient-based. This observation leads us to ask: Can we replace the inner loop trajectory collection and gradient update with a simpler, task-specific signal? In many practical settings, tasks are naturally accompanied by language instructions. Leveraging these instructions as a direct task-specific signal, we propose LA-MAML (Language Adapted MAML), which modifies the inner loop by adapting the global policy parameters in a single step through a learned embedding of the task instruction, replacing the inner loop trajectory collection and gradient-based updates. Experiments on the BabyAI benchmark demonstrate that LA-MAML achieves competitive or improved performance compared to baselines at a significantly lower per-iteration wall-clock training time. These results demonstrate that language instructions are an effective and efficient substitute for trajectory-based inner loop adaptation in meta RL.