This article introduces an imitation learning method for learning maximum entropy policies that comply with constraints demonstrated by expert trajectories executing a task. The formulation of the method takes advantage of results connecting performance to bounds for the KL-divergence between demonstrated and learned policies, and its objective is rigorously justified through a connection to a probabilistic inference framework for reinforcement learning, incorporating the reinforcement learning objective and the objective to abide by constraints in an entropy maximization setting. The proposed algorithm optimizes the learning objective with dual gradient descent, supporting effective and stable training. Experiments show that the proposed method can learn effective policy models for constraints-abiding behaviour, in settings with multiple constraints of different types, accommodating different modalities of demonstrated behaviour, and with abilities to generalize.

Imitation Learning from Observation (IfO) offers a powerful way to learn behaviors at large-scale: Unlike behavior cloning or offline reinforcement learning, IfO can leverage action-free demonstrations and thus circumvents the need for costly action-labeled demonstrations or reward functions. However, current IfO research focuses on idealized scenarios with mostly bimodal-quality data distributions, restricting the meaningfulness of the results. In contrast, this paper investigates more nuanced distributions and introduces a method to learn from such data, moving closer to a paradigm in which imitation learning can be performed iteratively via self-improvement. Our method adapts RL-based imitation learning to action-free demonstrations, using a value function to transfer information between expert and non-expert data. Through comprehensive evaluation, we delineate the relation between different data distributions and the applicability of algorithms and highlight the limitations of established methods. Our findings provide valuable insights for developing more robust and practical IfO techniques on a path to scalable behaviour learning.

Multi-task multi-agent reinforcement learning (MT-MARL) has recently gained attention for its potential to enhance MARL's adaptability across multiple tasks. However, it is challenging for existing multi-task learning methods to handle complex problems, as they are unable to handle unrelated tasks and possess limited knowledge transfer capabilities. In this paper, we propose a hierarchical approach that efficiently addresses these challenges. The high-level module utilizes a skill graph, while the low-level module employs a standard MARL algorithm. Our approach offers two contributions. First, we consider the MT-MARL problem in the context of unrelated tasks, expanding the scope of MTRL. Second, the skill graph is used as the upper layer of the standard hierarchical approach, with training independent of the lower layer, effectively handling unrelated tasks and enhancing knowledge transfer capabilities. Extensive experiments are conducted to validate these advantages and demonstrate that the proposed method outperforms the latest hierarchical MAPPO algorithms. Videos and code are available at https://github.com/WindyLab/MT-MARL-SG

Offline multi-task reinforcement learning aims to learn a unified policy capable of solving multiple tasks using only pre-collected task-mixed datasets, without requiring any online interaction with the environment. However, it faces significant challenges in effectively sharing knowledge across tasks. Inspired by the efficient knowledge abstraction observed in human learning, we propose Goal-Oriented Skill Abstraction (GO-Skill), a novel approach designed to extract and utilize reusable skills to enhance knowledge transfer and task performance. Our approach uncovers reusable skills through a goal-oriented skill extraction process and leverages vector quantization to construct a discrete skill library. To mitigate class imbalances between broadly applicable and task-specific skills, we introduce a skill enhancement phase to refine the extracted skills. Furthermore, we integrate these skills using hierarchical policy learning, enabling the construction of a high-level policy that dynamically orchestrates discrete skills to accomplish specific tasks. Extensive experiments on diverse robotic manipulation tasks within the MetaWorld benchmark demonstrate the effectiveness and versatility of GO-Skill.

Multi-task reinforcement learning endeavors to efficiently leverage shared information across various tasks, facilitating the simultaneous learning of multiple tasks. Existing approaches primarily focus on parameter sharing with carefully designed network structures or tailored optimization procedures. However, they overlook a direct and complementary way to exploit cross-task similarities: the control policies of tasks already proficient in some skills can provide explicit guidance for unmastered tasks to accelerate skills acquisition. To this end, we present a novel framework called Cross-Task Policy Guidance (CTPG), which trains a guide policy for each task to select the behavior policy interacting with the environment from all tasks' control policies, generating better training trajectories. In addition, we propose two gating mechanisms to improve the learning efficiency of CTPG: one gate filters out control policies that are not beneficial for guidance, while the other gate blocks tasks that do not necessitate guidance. CTPG is a general framework adaptable to existing parameter sharing approaches. Empirical evaluations demonstrate that incorporating CTPG with these approaches significantly enhances performance in manipulation and locomotion benchmarks.

Typically, learned robot controllers are trained via relatively unsystematic regimens and evaluated with coarse-grained outcome measures such as average cumulative reward. The typical approach is useful to compare learning algorithms but provides limited insight into the effects of different training regimens and little understanding about the richness and complexity of learned behaviors. Likewise, human infants and other animals are "trained" via unsystematic regimens, but in contrast, developmental psychologists evaluate their performance in highly-controlled experiments with fine-grained measures such as success, speed of walking, and prospective adjustments. However, the study of learned behavior in human infants is limited by the practical constraints of training and testing babies. Here, we present a case study that applies methods from developmental psychology to study the learned behavior of the simulated bipedal robot Cassie. Following research on infant walking, we systematically designed reinforcement learning training regimens and tested the resulting controllers in simulated environments analogous to those used for babies--but without the practical constraints. Results reveal new insights into the behavioral impact of different training regimens and the development of Cassie's learned behaviors relative to infants who are learning to walk. This interdisciplinary baby-robot approach provides inspiration for future research designed to systematically test effects of training on the development of complex learned robot behaviors.

Deep brain stimulation (DBS) is an established intervention for Parkinson's disease (PD), but conventional open-loop systems lack adaptability, are energy-inefficient due to continuous stimulation, and provide limited personalization to individual neural dynamics. Adaptive DBS (aDBS) offers a closed-loop alternative, using biomarkers such as beta-band oscillations to dynamically modulate stimulation. While reinforcement learning (RL) holds promise for personalized aDBS control, existing methods suffer from high sample complexity, unstable exploration in binary action spaces, and limited deployability on resource-constrained hardware. We propose SEA-DBS, a sample-efficient actor-critic framework that addresses the core challenges of RL-based adaptive neurostimulation. SEA-DBS integrates a predictive reward model to reduce reliance on real-time feedback and employs Gumbel Softmax-based exploration for stable, differentiable policy updates in binary action spaces. Together, these components improve sample efficiency, exploration robustness, and compatibility with resource-constrained neuromodulatory hardware. We evaluate SEA-DBS on a biologically realistic simulation of Parkinsonian basal ganglia activity, demonstrating faster convergence, stronger suppression of pathological beta-band power, and resilience to post-training FP16 quantization. Our results show that SEA-DBS offers a practical and effective RL-based aDBS framework for real-time, resource-constrained neuromodulation.

Humans can pursue a near-infinite variety of tasks, but typically can only pursue a small number at the same time. We hypothesize that humans leverage experience on one task to preemptively learn solutions to other tasks that were accessible but not pursued. We formalize this idea as Multitask Preplay, a novel algorithm that replays experience on one task as the starting point for "preplay" -- counterfactual simulation of an accessible but unpursued task. Preplay is used to learn a predictive representation that can support fast, adaptive task performance later on. We first show that, compared to traditional planning and predictive representation methods, multitask preplay better predicts how humans generalize to tasks that were accessible but not pursued in a small grid-world, even when people didn't know they would need to generalize to these tasks. We then show these predictions generalize to Craftax, a partially observable 2D Minecraft environment. Finally, we show that Multitask Preplay enables artificial agents to learn behaviors that transfer to novel Craftax worlds sharing task co-occurrence structure. These findings demonstrate that Multitask Preplay is a scalable theory of how humans counterfactually learn and generalize across multiple tasks; endowing artificial agents with the same capacity can significantly improve their performance in challenging multitask environments.

For effective real-world deployment, robots should adapt to human preferences, such as balancing distance, time, and safety in delivery routing. Active preference learning (APL) learns human reward functions by presenting trajectories for ranking. However, existing methods often struggle to explore the full trajectory space and fail to identify informative queries, particularly in long-horizon tasks. We propose CRED, a trajectory generation method for APL that improves reward estimation by jointly optimizing environment design and trajectory selection. CRED "imagines" new scenarios through environment design and uses counterfactual reasoning -- by sampling rewards from its current belief and asking "What if this reward were the true preference?" -- to generate a diverse and informative set of trajectories for ranking. Experiments in GridWorld and real-world navigation using OpenStreetMap data show that CRED improves reward learning and generalizes effectively across different environments.

Training large language models with reinforcement learning (RL) against verifiable rewards significantly enhances their reasoning abilities, yet remains computationally expensive due to inefficient uniform prompt sampling. We introduce Selective Prompting with Efficient Estimation of Difficulty (SPEED), an adaptive online RL curriculum that selectively chooses training examples of intermediate difficulty to maximize learning efficiency. Theoretically, we establish that intermediate-difficulty prompts improve the gradient estimator's signal-to-noise ratio, accelerating convergence. Empirically, our efficient implementation leads to 2x to 6x faster training without degrading accuracy, requires no manual tuning, and integrates seamlessly into standard RL algorithms.