DISCOVER: Automated Curricula for Sparse-Reward Reinforcement Learning

Sparse-reward reinforcement learning (RL) can model a wide range of highly complex tasks. Solving sparse-reward tasks is RL's core premise - requiring efficient exploration coupled with long-horizon credit assignment - and overcoming these challenges is key for building self-improving agents with superhuman ability. We argue that solving complex and high-dimensional tasks requires solving simpler tasks that are relevant to the target task. In contrast, most prior work designs strategies for selecting exploratory tasks with the objective of solving any task, making exploration of challenging high-dimensional, long-horizon tasks intractable. We find that the sense of direction, necessary for effective exploration, can be extracted from existing RL algorithms, without needing any prior information. Based on this finding, we propose a method for directed sparse-reward goal-conditioned very long-horizon RL (DISCOVER), which selects exploratory goals in the direction of the target task. We connect DISCOVER to principled exploration in bandits, formally bounding the time until the target task becomes achievable in terms of the agent's initial distance to the target, but independent of the volume of the space of all tasks. Empirically, we perform a thorough evaluation in high-dimensional environments. We find that the directed goal selection of DISCOVER solves exploration problems that are beyond the reach of prior state-of-the-art exploration methods in RL.

Local Mixtures of Experts: Essentially Free Test-Time Training via Model Merging

Mixture of expert (MoE) models are a promising approach to increasing model capacity without increasing inference cost, and are core components of many state-of-the-art language models. However, current MoE models typically use only few experts due to prohibitive training and inference cost. We propose Test-Time Model Merging (TTMM) which scales the MoE paradigm to an order of magnitude more experts and uses model merging to avoid almost any test-time overhead. We show that TTMM is an approximation of test-time training (TTT), which fine-tunes an expert model for each prediction task, i.e., prompt. TTT has recently been shown to significantly improve language models, but is computationally expensive. We find that performance of TTMM improves with more experts and approaches the performance of TTT. Moreover, we find that with a 1B parameter base model, TTMM is more than 100x faster than TTT at test-time by amortizing the cost of TTT at train-time. Thus, TTMM offers a promising cost-effective approach to scale test-time training.

Finite-Sample-Based Reachability for Safe Control with Gaussian Process Dynamics

Gaussian Process (GP) regression is shown to be effective for learning unknown dynamics, enabling efficient and safety-aware control strategies across diverse applications. However, existing GP-based model predictive control (GP-MPC) methods either rely on approximations, thus lacking guarantees, or are overly conservative, which limits their practical utility. To close this gap, we present a sampling-based framework that efficiently propagates the model's epistemic uncertainty while avoiding conservatism. We establish a novel sample complexity result that enables the construction of a reachable set using a finite number of dynamics functions sampled from the GP posterior. Building on this, we design a sampling-based GP-MPC scheme that is recursively feasible and guarantees closed-loop safety and stability with high probability. Finally, we showcase the effectiveness of our method on two numerical examples, highlighting accurate reachable set over-approximation and safe closed-loop performance.

Offline Robotic World Model: Learning Robotic Policies without a Physics Simulator

Reinforcement Learning (RL) has demonstrated impressive capabilities in robotic control but remains challenging due to high sample complexity, safety concerns, and the sim-to-real gap. While offline RL eliminates the need for risky real-world exploration by learning from pre-collected data, it suffers from distributional shift, limiting policy generalization. Model-Based RL (MBRL) addresses this by leveraging predictive models for synthetic rollouts, yet existing approaches often lack robust uncertainty estimation, leading to compounding errors in offline settings. We introduce Offline Robotic World Model (RWM-O), a model-based approach that explicitly estimates epistemic uncertainty to improve policy learning without reliance on a physics simulator. By integrating these uncertainty estimates into policy optimization, our approach penalizes unreliable transitions, reducing overfitting to model errors and enhancing stability. Experimental results show that RWM-O improves generalization and safety, enabling policy learning purely from real-world data and advancing scalable, data-efficient RL for robotics.

Ride-Sourcing Vehicle Rebalancing with Service Accessibility Guarantees via Constrained Mean-Field Reinforcement Learning

The rapid expansion of ride-sourcing services such as Uber, Lyft, and Didi Chuxing has fundamentally reshaped urban transportation by offering flexible, on-demand mobility via mobile applications. Despite their convenience, these platforms confront significant operational challenges, particularly vehicle rebalancing - the strategic repositioning of thousands of vehicles to address spatiotemporal mismatches in supply and demand. Inadequate rebalancing results in prolonged rider waiting times, inefficient vehicle utilization, and inequitable distribution of services, leading to disparities in driver availability and income. To tackle these complexities, we introduce scalable continuous-state mean-field control (MFC) and reinforcement learning (MFRL) models that explicitly represent each vehicle's precise location and employ continuous repositioning actions guided by the distribution of other vehicles. To ensure equitable service distribution, an accessibility constraint is integrated within our optimal control formulation, balancing operational efficiency with equitable access to the service across geographic regions. Our approach acknowledges realistic conditions, including inherent stochasticity in transitions, the simultaneous occurrence of vehicle-rider matching, vehicles' rebalancing and cruising, and variability in rider behaviors. Crucially, we relax the traditional mean-field assumption of equal supply-demand volume, better reflecting practical scenarios. Extensive empirical evaluation using real-world data-driven simulation of Shenzhen demonstrates the real-time efficiency and robustness of our approach at the scale of tens of thousands of vehicles. The code is available at https://github.com/mjusup1501/mf-vehicle-rebalancing.

NIL: No-data Imitation Learning by Leveraging Pre-trained Video Diffusion Models

Acquiring physically plausible motor skills across diverse and unconventional morphologies-including humanoid robots, quadrupeds, and animals-is essential for advancing character simulation and robotics. Traditional methods, such as reinforcement learning (RL) are task- and body-specific, require extensive reward function engineering, and do not generalize well. Imitation learning offers an alternative but relies heavily on high-quality expert demonstrations, which are difficult to obtain for non-human morphologies. Video diffusion models, on the other hand, are capable of generating realistic videos of various morphologies, from humans to ants. Leveraging this capability, we propose a data-independent approach for skill acquisition that learns 3D motor skills from 2D-generated videos, with generalization capability to unconventional and non-human forms. Specifically, we guide the imitation learning process by leveraging vision transformers for video-based comparisons by calculating pair-wise distance between video embeddings. Along with video-encoding distance, we also use a computed similarity between segmented video frames as a guidance reward. We validate our method on locomotion tasks involving unique body configurations. In humanoid robot locomotion tasks, we demonstrate that 'No-data Imitation Learning' (NIL) outperforms baselines trained on 3D motion-capture data. Our results highlight the potential of leveraging generative video models for physically plausible skill learning with diverse morphologies, effectively replacing data collection with data generation for imitation learning.

Performance-driven Constrained Optimal Auto-Tuner for MPC

A key challenge in tuning Model Predictive Control (MPC) cost function parameters is to ensure that the system performance stays consistently above a certain threshold. To address this challenge, we propose a novel method, COAT-MPC, Constrained Optimal Auto-Tuner for MPC. With every tuning iteration, COAT-MPC gathers performance data and learns by updating its posterior belief. It explores the tuning parameters' domain towards optimistic parameters in a goal-directed fashion, which is key to its sample efficiency. We theoretically analyze COAT-MPC, showing that it satisfies performance constraints with arbitrarily high probability at all times and provably converges to the optimum performance within finite time. Through comprehensive simulations and comparative analyses with a hardware platform, we demonstrate the effectiveness of COAT-MPC in comparison to classical Bayesian Optimization (BO) and other state-of-the-art methods. When applied to autonomous racing, our approach outperforms baselines in terms of constraint violations and cumulative regret over time.

Confidence Estimation via Sequential Likelihood Mixing

We present a universal framework for constructing confidence sets based on sequential likelihood mixing. Building upon classical results from sequential analysis, we provide a unifying perspective on several recent lines of work, and establish fundamental connections between sequential mixing, Bayesian inference and regret inequalities from online estimation. The framework applies to any realizable family of likelihood functions and allows for non-i.i.d. data and anytime validity. Moreover, the framework seamlessly integrates standard approximate inference techniques, such as variational inference and sampling-based methods, and extends to misspecified model classes, while preserving provable coverage guarantees. We illustrate the power of the framework by deriving tighter confidence sequences for classical settings, including sequential linear regression and sparse estimation, with simplified proofs.

Probabilistic Artificial Intelligence

Artificial intelligence commonly refers to the science and engineering of artificial systems that can carry out tasks generally associated with requiring aspects of human intelligence, such as playing games, translating languages, and driving cars. In recent years, there have been exciting advances in learning-based, data-driven approaches towards AI, and machine learning and deep learning have enabled computer systems to perceive the world in unprecedented ways. Reinforcement learning has enabled breakthroughs in complex games such as Go and challenging robotics tasks such as quadrupedal locomotion. A key aspect of intelligence is to not only make predictions, but reason about the uncertainty in these predictions, and to consider this uncertainty when making decisions. This is what this manuscript on "Probabilistic Artificial Intelligence" is about. The first part covers probabilistic approaches to machine learning. We discuss the differentiation between "epistemic" uncertainty due to lack of data and "aleatoric" uncertainty, which is irreducible and stems, e.g., from noisy observations and outcomes. We discuss concrete approaches towards probabilistic inference and modern approaches to efficient approximate inference. The second part of the manuscript is about taking uncertainty into account in sequential decision tasks. We consider active learning and Bayesian optimization -- approaches that collect data by proposing experiments that are informative for reducing the epistemic uncertainty. We then consider reinforcement learning and modern deep RL approaches that use neural network function approximation. We close by discussing modern approaches in model-based RL, which harness epistemic and aleatoric uncertainty to guide exploration, while also reasoning about safety.

Toward Task Generalization via Memory Augmentation in Meta-Reinforcement Learning

In reinforcement learning (RL), agents often struggle to perform well on tasks that differ from those encountered during training. This limitation presents a challenge to the broader deployment of RL in diverse and dynamic task settings. In this work, we introduce memory augmentation, a memory-based RL approach to improve task generalization. Our approach leverages task-structured augmentations to simulate plausible out-of-distribution scenarios and incorporates memory mechanisms to enable context-aware policy adaptation. Trained on a predefined set of tasks, our policy demonstrates the ability to generalize to unseen tasks through memory augmentation without requiring additional interactions with the environment. Through extensive simulation experiments and real-world hardware evaluations on legged locomotion tasks, we demonstrate that our approach achieves zero-shot generalization to unseen tasks while maintaining robust in-distribution performance and high sample efficiency.