Offline reinforcement learning (RL) learns policies from fixed datasets without online interactions, but suffers from distribution shift, causing inaccurate evaluation and overestimation of out-of-distribution (OOD) actions. Existing methods counter this by conservatively discouraging all OOD actions, which limits generalization. We propose Advantage-based Diffusion Actor-Critic (ADAC), which evaluates OOD actions via an advantage-like function and uses it to modulate the Q-function update discriminatively. Our key insight is that the (state) value function is generally learned more reliably than the action-value function; we thus use the next-state value to indirectly assess each action. We develop a PointMaze environment to clearly visualize that advantage modulation effectively selects superior OOD actions while discouraging inferior ones. Moreover, extensive experiments on the D4RL benchmark show that ADAC achieves state-of-the-art performance, with especially strong gains on challenging tasks.

Multi-objective optimization problems, which require the simultaneous optimization of multiple objectives, are prevalent across numerous applications. Existing multi-objective optimization methods often rely on manually-tuned aggregation functions to formulate a joint optimization objective. The performance of such hand-tuned methods is heavily dependent on careful weight selection, a time-consuming and laborious process. These limitations also arise in the setting of reinforcement-learning-based motion tracking methods for physically simulated characters, where intricately crafted reward functions are typically used to achieve high-fidelity results. Such solutions not only require domain expertise and significant manual tuning, but also limit the applicability of the resulting reward function across diverse skills. To bridge this gap, we present a novel adversarial multi-objective optimization technique that is broadly applicable to a range of multi-objective reinforcement-learning tasks, including motion tracking. Our proposed Adversarial Differential Discriminator (ADD) receives a single positive sample, yet is still effective at guiding the optimization process. We demonstrate that our technique can enable characters to closely replicate a variety of acrobatic and agile behaviors, achieving comparable quality to state-of-the-art motion-tracking methods, without relying on manually-designed reward functions. Code and results are available at https://add-moo.github.io/.

Inferring an adversary's goals from exhibited behavior is crucial for counterplanning and non-cooperative multi-agent systems in domains like cybersecurity, military, and strategy games. Deep Inverse Reinforcement Learning (IRL) methods based on maximum entropy principles show promise in recovering adversaries' goals but are typically offline, require large batch sizes with gradient descent, and rely on first-order updates, limiting their applicability in real-time scenarios. We propose an online Recursive Deep Inverse Reinforcement Learning (RDIRL) approach to recover the cost function governing the adversary actions and goals. Specifically, we minimize an upper bound on the standard Guided Cost Learning (GCL) objective using sequential second-order Newton updates, akin to the Extended Kalman Filter (EKF), leading to a fast (in terms of convergence) learning algorithm. We demonstrate that RDIRL is able to recover cost and reward functions of expert agents in standard and adversarial benchmark tasks. Experiments on benchmark tasks show that our proposed approach outperforms several leading IRL algorithms.

Offline reinforcement learning (RL) learns effective policies from pre-collected datasets, offering a practical solution for applications where online interactions are risky or costly. Model-based approaches are particularly advantageous for offline RL, owing to their data efficiency and generalizability. However, due to inherent model errors, model-based methods often artificially introduce conservatism guided by heuristic uncertainty estimation, which can be unreliable. In this paper, we introduce VIPO, a novel model-based offline RL algorithm that incorporates self-supervised feedback from value estimation to enhance model training. Specifically, the model is learned by additionally minimizing the inconsistency between the value learned directly from the offline data and the one estimated from the model. We perform comprehensive evaluations from multiple perspectives to show that VIPO can learn a highly accurate model efficiently and consistently outperform existing methods. In particular, it achieves state-of-the-art performance on almost all tasks in both D4RL and NeoRL benchmarks. Overall, VIPO offers a general framework that can be readily integrated into existing model-based offline RL algorithms to systematically enhance model accuracy.

We introduce magnitude and direction (MAD) policies, a policy parameterization for reinforcement learning (RL) that preserves Lp closed-loop stability for nonlinear dynamical systems. Despite their completeness in describing all stabilizing controllers, methods based on nonlinear Youla and system-level synthesis are significantly impacted by the difficulty of parametrizing Lp-stable operators. In contrast, MAD policies introduce explicit feedback on state-dependent features - a key element behind the success of reinforcement learning pipelines - without jeopardizing closed-loop stability. This is achieved by letting the magnitude of the control input be described by a disturbance-feedback Lp-stable operator, while selecting its direction based on state-dependent features through a universal function approximator. We further characterize the robust stability properties of MAD policies under model mismatch. Unlike existing disturbance-feedback policy parametrizations, MAD policies introduce state-feedback components compatible with model-free RL pipelines, ensuring closed-loop stability with no model information beyond assuming open-loop stability. Numerical experiments show that MAD policies trained with deep deterministic policy gradient (DDPG) methods generalize to unseen scenarios - matching the performance of standard neural network policies while guaranteeing closed-loop stability by design.

Robotic locomotion research typically draws from biologically inspired leg designs, yet many human-engineered settings can benefit from non-anthropomorphic forms. TARS3D translates the block-shaped 'TARS' robot from Interstellar into a 0.25 m, 0.99 kg research platform with seven actuated degrees of freedom. The film shows two primary gaits: a bipedal-like walk and a high-speed rolling mode. For TARS3D, we build reduced-order models for each, derive closed-form limit-cycle conditions, and validate the predictions on hardware. Experiments confirm that the robot respects its +/-150 degree hip limits, alternates left-right contacts without interference, and maintains an eight-step hybrid limit cycle in rolling mode. Because each telescopic leg provides four contact corners, the rolling gait is modeled as an eight-spoke double rimless wheel. The robot's telescopic leg redundancy implies a far richer gait repertoire than the two limit cycles treated analytically. So, we used deep reinforcement learning (DRL) in simulation to search the unexplored space. We observed that the learned policy can recover the analytic gaits under the right priors and discover novel behaviors as well. Our findings show that TARS3D's fiction-inspired bio-transcending morphology can realize multiple previously unexplored locomotion modes and that further learning-driven search is likely to reveal more. This combination of analytic synthesis and reinforcement learning opens a promising pathway for multimodal robotics.

Large Reasoning Models (LRMs) often exhibit a tendency for overanalysis in simple tasks, where the models excessively utilize System 2-type, deliberate reasoning, leading to inefficient token generation. Furthermore, these models face challenges in adapting their reasoning capabilities to rapidly changing environments due to the static nature of their pretraining data. To address these issues, advancing Large Language Models (LLMs) for complex reasoning tasks requires innovative approaches that bridge intuitive and deliberate cognitive processes, akin to human cognition's dual-system dynamic. This paper introduces a Multi-Agent System for Deep ReSearch (MARS) enabling seamless integration of System 1's fast, intuitive thinking with System 2's deliberate reasoning within LLMs. MARS strategically integrates multiple external tools, such as Google Search, Google Scholar, and Python Interpreter, to access up-to-date information and execute complex computations, while creating a specialized division of labor where System 1 efficiently processes and summarizes high-volume external information, providing distilled insights that expand System 2's reasoning context without overwhelming its capacity. Furthermore, we propose a multi-agent reinforcement learning framework extending Group Relative Policy Optimization to simultaneously optimize both systems with multi-turn tool interactions, bin-packing optimization, and sample balancing strategies that enhance collaborative efficiency. Extensive experiments demonstrate MARS achieves substantial improvements of 3.86% on the challenging Humanity's Last Exam (HLE) benchmark and an average gain of 8.9% across 7 knowledge-intensive tasks, validating the effectiveness of our dual-system paradigm for complex reasoning in dynamic information environments.

Skills are essential for unlocking higher levels of problem solving. A common approach to discovering these skills is to learn ones that reliably reach different states, thus empowering the agent to control its environment. However, existing skill discovery algorithms often overlook the natural state variables present in many reinforcement learning problems, meaning that the discovered skills lack control of specific state variables. This can significantly hamper exploration efficiency, make skills more challenging to learn with, and lead to negative side effects in downstream tasks when the goal is under-specified. We introduce a general method that enables these skill discovery algorithms to learn focused skills -- skills that target and control specific state variables. Our approach improves state space coverage by a factor of three, unlocks new learning capabilities, and automatically avoids negative side effects in downstream tasks.

Humans are good at learning on the job: We learn how to solve the tasks we face as we go along. Can a model do the same? We propose an agent that assembles a task-specific curriculum, called test-time curriculum (TTC-RL), and applies reinforcement learning to continue training the model for its target task. The test-time curriculum avoids time-consuming human curation of datasets by automatically selecting the most task-relevant data from a large pool of available training data. Our experiments demonstrate that reinforcement learning on a test-time curriculum consistently improves the model on its target tasks, across a variety of evaluations and models. Notably, on challenging math and coding benchmarks, TTC-RL improves the pass@1 of Qwen3-8B by approximately 1.8x on AIME25 and 2.1x on CodeElo. Moreover, we find that TTC-RL significantly raises the performance ceiling compared to the initial model, increasing pass@8 on AIME25 from 40% to 62% and on CodeElo from 28% to 43%. Our findings show the potential of test-time curricula in extending the test-time scaling paradigm to continual training on thousands of task-relevant experiences during test-time.

Enabling large language models (LLMs) to utilize search tools offers a promising path to overcoming fundamental limitations such as knowledge cutoffs and hallucinations. Recent work has explored reinforcement learning (RL) for training search-augmented agents that interleave reasoning and retrieval before answering. These approaches usually rely on outcome-based rewards (e.g., exact match), implicitly assuming that optimizing for final answers will also yield effective intermediate search behaviors. Our analysis challenges this assumption: we uncover multiple systematic deficiencies in search that arise under outcome-only training and ultimately degrade final answer quality, including failure to invoke tools, invalid queries, and redundant searches. To address these shortcomings, we introduce DeSA (Decoupling Search-and-Answering), a simple two-stage training framework that explicitly separates search optimization from answer generation. In Stage 1, agents are trained to improve search effectiveness with retrieval recall-based rewards. In Stage 2, outcome rewards are employed to optimize final answer generation. Across seven QA benchmarks, DeSA-trained agents consistently improve search behaviors, delivering substantially higher search recall and answer accuracy than outcome-only baselines. Notably, DeSA outperforms single-stage training approaches that simultaneously optimize recall and outcome rewards, underscoring the necessity of explicitly decoupling the two objectives.